Planet FSCI

June 16, 2026

• Thejeshgn GN Development as Freedom

  This time, my trip to Chennai was also a road trip. I got a ride with friends in the car, and it also gave us lots of time for conversation. As we were going to the convocation, the discussions turned to education and degrees. Though there are lots of folks dissing college education and degrees in the tech world. I thought that a college education and formal degrees have value. And to know its value, one should speak to someone who is the first graduate in the family and what kind of changes it brings.

  I think we all kind of agreed, and friend mentioned an article by Amartya Sen on education and human capability. I did find myself agreeing with the points and, hence, searched for that article, which led me to the book, Development as Freedom. I am reading it currently. I think it’s a valuable read.

  Also this take on the book by Denis O’Hearn on it’s tenth year of publication.

  The center of Sen’s vision is what he calls a ‘capability approach’, where the basic concern of human development is ‘our capability to lead the kind of lives we have reason to value’, rather than the usual concentration on rising GDP, technical progress, or industrialization (Sen, 1999:285). His approach ‘inescapably focuses on the agency and judgment of individuals’ (ibid.:288) including their capability, responsibility, and opportunity. Raising human capability is good because it improves: the choices, wellbeing, and freedom of people; their role in influencing social change; and their role in influencing economic production.

  He painstakingly distinguishes human capability from human capital. Human capital is important, as it refers to the agency of people in augmenting production possibilities. Yet human capability is more important because it refers to the substantive freedom of people to lead the lives they have reason to value and to enhance the real choices they have. Education, for example, is crucial beyond its role in production; its most important role being that of increasing human capability and therefore choice. Again, Sen cites Adam Smith who links productive abilities to lifestyles to education and training, and presumes the improvability of each. While the popularity of the concept of human capital is for Sen ‘certainly an enriching move’, it needs supplementation by an approach that takes human capability as its central concern.

  Read the book if you have time.

  ---

  You can read this blog using RSS Feed. But if you are the person who loves getting emails, then you can join my readers by signing up.

  Subscribe

---

June 13, 2026

• Thejeshgn GN Back to Basics: Build Your Own LLM from Scratch

  I did a workshop titled “Back to Basics: Build Your Own LLM from Scratch” at IITM/Paradox 2026, which kind of included some basic theory on how a transformer works, and then building a very small LLM. The idea was to demystify an LLM (or transformer) by understanding what goes on and then building one to deepen our understanding. I had to skip some slides because the planned session was only two hours. Ideally, I want it to be around 4 hours, split into 2 sessions: one for theory and one for lab. Maybe next time, when I plan, I will make it 4 hours so I can do it at a slower pace.

  Of course, there are other similar workshops available online, and some of them are linked in the references section. This is just my take on it and what I used for my own understanding.

  Suppose you want to try it at your own pace. Try the slides below and then use annotated code to read and run. Slides and code are in a repo (CB, GH) too if you prefer that.

  

  #!/usr/bin/env -S uv run --script
  # /// script
  # requires-python = ">=3.10"
  # dependencies = [
  #   "torch==2.8.0",
  # ]
  # [tool.uv]
  # extra-index-url = ["https://download.pytorch.org/whl/cpu"]
  # ///
  """
  build_and_test.py  (ANNOTATED VERSION FOR WORKSHOP)
  A minimal, single-file GPT for "Back to Basics: Build Your Own LLM from Scratch".
  
  �������������������������������������������������������������������������������
  WALKTHROUGH MAP — suggested order
  �������������������������������������������������������������������������������
  
    ① GPTConfig            � slide "The whole model in 5 numbers"
    ② CharTokenizer        � slides "Tokens" / "Tokenizer"
    ③ GPT.__init__/forward � slide "Recap: what we just built" (the big pipeline)
    ④ CausalSelfAttention  � slides "Single-head attention" → "Combining heads with Wo"
    ⑤ FeedForward          � slide "Feed-Forward Network (FFN)"
    ⑥ TransformerBlock     � slide "One full transformer block"
    ⑦ get_batch            � (where x/y "next-token" pairs come from)
    ⑧ train()              � slides "Cross-entropy" → "Training"
    ⑨ GPT.generate()       � slide "Generation: from logits to text"
  
  Comments marked 💬 are things worth SAYING out loud.
  Comments marked � are good questions to ASK the room.
  Comments marked ⚠� are common gotchas / likely audience questions.
  
  Usage:
      # Train on a text file (CPU only, by design)
      uv run build_and_test.py train --data ../data/shakespeare.txt --max-steps 2000
  
      # Generate from a saved checkpoint
      uv run build_and_test.py generate --checkpoint ../checkpoints/run1/final_checkpoint.pt --prompt "To be, or not " --num-new-tokens 200 --temperature 0.8 --top-k 40 --seed 42
  """
  
  import argparse
  import csv
  import math
  import os
  import sys
  import time
  from dataclasses import dataclass, asdict
  
  import torch
  import torch.nn as nn
  import torch.nn.functional as F
  
  
  # �����������������������������������������������������������������������������
  # â‘  CONFIG                                  [Slide: "The whole model in 5 numbers"]
  # �����������������������������������������������������������������������������
  # 💬 "These five numbers ARE the model. Everything else is derived from them."
  #    Point back to this class every time a shape like (B, T, C) appears below:
  #       B = batch_size, T ≤ block_size, C = n_embd.
  
  @dataclass
  class GPTConfig:
      # Architecture (the 5 numbers from the slides)
      vocab_size: int = 65        # how many unique tokens (set from data, after tokenizer)
      n_embd:     int = 256       # C: the vector size each token is represented by
      n_head:     int = 4         # parallel attention heads (d_k = n_embd/n_head = 64)
      block_size: int = 256       # T_max: the longest context the model can ever see
      n_layer:    int = 4         # how many TransformerBlocks we stack
  
  # Training knobs — deliberately NOT in GPTConfig:
  # 💬 "These shape the *training run*, not the *model*. A checkpoint doesn't need them."
  batch_size = 32   # sequences per training step (the B in (B, T, C))
  dropout = 0.1     # ⚠� on during training, automatically off in eval()
                    #    [Slide: "Training vs. inference" — dropout row]
  
  output_path = "../checkpoints/run1"
  
  # �����������������������������������������������������������������������������
  # â‘¡ TOKENIZER                                    [Slides: "Tokens" / "Tokenizer"]
  # �����������������������������������������������������������������������������
  # 💬 "A tokenizer has exactly two jobs: encode (text → IDs) and decode (IDs → text).
  #    We picked character-level — the simplest of the three choices on the slide.
  #    GPT/LLaMA/Claude use subword BPE; same idea, fancier vocab."
  #
  # � Ask: "If our vocab is the 65 unique characters in Shakespeare, what happens
  #    when you prompt with an emoji?" → see encode(): it silently drops unknowns.
  
  class CharTokenizer:
      """Smallest possible tokenizer: one character = one token."""
  
      def __init__(self, vocab: list[str]):
          self.vocab = vocab
          # stoi = "string to int", itos = "int to string" — two dicts, that's it.
          self.stoi = {ch: i for i, ch in enumerate(vocab)}
          self.itos = {i: ch for i, ch in enumerate(vocab)}
  
      @classmethod
      def from_text(cls, text: str) -> "CharTokenizer":
          # 💬 "The vocab is just every unique character that occurs in the data."
          # Sorted so the vocab is deterministic across runs
          # ⚠� Without sorted(), set() ordering varies → token IDs change between runs
          #    → an old checkpoint would decode to garbage. This one line is why we
          #    can reload checkpoints reliably.
          vocab = sorted(list(set(text)))
          return cls(vocab)
  
      def encode(self, s: str) -> list[int]:
          # text → list of integers.  [Slide: "Tokenizer" — encode example]
          # `if c in self.stoi`: characters not in the training data are dropped.
          return [self.stoi[c] for c in s if c in self.stoi]
  
      def decode(self, ids: list[int]) -> str:
          # integers → text. Perfect inverse of encode (for known chars).
          return "".join(self.itos[i] for i in ids)
  
      @property
      def vocab_size(self) -> int:
          # 💬 "This becomes the first of our 5 numbers — vocab_size in GPTConfig."
          return len(self.vocab)
  
  
  # �����������������������������������������������������������������������������
  # ④ ATTENTION             [Slides: "Why attention?" → "Combining heads with Wo"]
  # �����������������������������������������������������������������������������
  # 💬 "This class is the heart of the workshop. The 7 numbered steps in forward()
  #    map one-to-one onto the attention slides. Everything else is plumbing."
  #
  # Teaching tip: walk forward() with a concrete shape, e.g. B=32, T=256, C=256,
  # n_head=4, d_k=64 — and write the shapes on the board as you go.
  
  class CausalSelfAttention(nn.Module):
      """Multi-head causal self-attention. One linear projects to Q,K,V together."""
  
      def __init__(self, cfg: GPTConfig):
          super().__init__()
          # d_k = n_embd / n_head must divide evenly — each head gets a clean slice.
          # [Slide: "Multi-head attention" — d_k = n_embd / n_head]
          assert cfg.n_embd % cfg.n_head == 0, "n_embd must be divisible by n_head"
          self.n_head = cfg.n_head
          self.n_embd = cfg.n_embd
          self.d_k = cfg.n_embd // cfg.n_head
  
          # 💬 "The slides show three separate matrices Wq, Wk, Wv, each (C × C).
          #    In code we fuse them into ONE (C × 3C) matrix for efficiency —
          #    one matmul instead of three. Same math, same parameter count."
          # [Slide: "Single-head attention: Q, K, V"]
          self.qkv = nn.Linear(cfg.n_embd, 3 * cfg.n_embd)
  
          # Wo from the slides: lets the heads talk to each other after running
          # independently. Without it, heads would be siloed.
          # [Slide: "Combining heads with Wo"]
          self.proj = nn.Linear(cfg.n_embd, cfg.n_embd)
          self.dropout = nn.Dropout(dropout)
  
          # 💬 "The causal mask is NOT learned — it's a fixed triangle of 1s.
          #    Row i has 1s up to column i: 'token i may look at tokens 0..i'."
          # [Slide: "Causal mask"] and [Slide: "What gets learned, what stays fixed"]
          # register_buffer = "part of the model, moves with .to(device),
          # saved in state_dict, but NO gradients" — perfect for a constant.
          mask = torch.tril(torch.ones(cfg.block_size, cfg.block_size))
          self.register_buffer("mask", mask.view(1, 1, cfg.block_size, cfg.block_size))
  
      def forward(self, x):
          B, T, C = x.shape  # batch, seq_len, n_embd — write these on the board
  
          # ── 1) Project to Q, K, V ──────────────── [Slide: "Q, K, V"]
          # One big matmul gives (B, T, 3C); split() carves it into three (B, T, C).
          # 💬 "Query: what am I looking for? Key: what do I offer? Value: what do
          #    I pass along if matched?"
          q, k, v = self.qkv(x).split(self.n_embd, dim=2)
  
          # ── 2) Split into heads ────────── [Slide: "Multi-head attention"]
          # (B, T, C) → (B, T, n_head, d_k) → transpose → (B, n_head, T, d_k)
          # 💬 "No new computation here — we're just reshaping so each head can run
          #    the SAME attention math independently on its own d_k-sized slice."
          q = q.view(B, T, self.n_head, self.d_k).transpose(1, 2)
          k = k.view(B, T, self.n_head, self.d_k).transpose(1, 2)
          v = v.view(B, T, self.n_head, self.d_k).transpose(1, 2)
  
          # ── 3) Scaled dot-product scores ──── [Slide: "Attention scores (scaled)"]
          # (B, nh, T, d_k) @ (B, nh, d_k, T) → (B, nh, T, T)
          # 💬 "A T×T grid per head: how relevant is every token to every other token."
          # ⚠� The 1/√d_k scaling is the line students forget. Without it, dot
          #    products grow with d_k → softmax saturates → gradients vanish.
          scores = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(self.d_k))
  
          # ── 4) Causal mask ──────────────────────── [Slide: "Causal mask"]
          # Where the triangle has 0 (future positions), drop in -inf.
          # 💬 "-inf BEFORE softmax becomes exactly 0 AFTER softmax — the model
          #    literally cannot peek at the answer."
          # [:T, :T] crops the precomputed block_size mask to the actual seq length.
          scores = scores.masked_fill(self.mask[:, :, :T, :T] == 0, float("-inf"))
  
          # ── 5) Softmax → attention weights ──── [Slide: "Softmax intuition"]
          # Each row becomes a probability distribution: positive, sums to 1,
          # behaves like "importance".
          attn = F.softmax(scores, dim=-1)
          attn = self.dropout(attn)  # regularization: randomly drop some attention links
  
          # ── 6) Apply attention to V ────────────── [Slide: "Apply attention"]
          # (B, nh, T, T) @ (B, nh, T, d_k) → (B, nh, T, d_k)
          # 💬 "Each output row is a weighted BLEND of value vectors from earlier
          #    positions. THIS is the heart of the transformer."
          out = attn @ v
  
          # ── 7) Re-combine heads + Wo ──── [Slide: "Combining heads with Wo"]
          # (B, nh, T, d_k) → (B, T, nh, d_k) → (B, T, C): concat of all heads.
          # ⚠� .contiguous() is needed because transpose only changes the view,
          #    not memory layout — .view() requires contiguous memory.
          out = out.transpose(1, 2).contiguous().view(B, T, C)
          out = self.proj(out)  # Wo: mixes information across heads
          return out
  
  
  # �����������������������������������������������������������������������������
  # ⑤ FFN                                  [Slide: "Feed-Forward Network (FFN)"]
  # �����������������������������������������������������������������������������
  # 💬 "Attention mixes information ACROSS tokens; the FFN processes each token
  #    INDEPENDENTLY — same two layers applied to every position. Expand to 4×
  #    (room to think), non-linearity, compress back."
  
  class FeedForward(nn.Module):
      """Two-layer MLP: expand to 4x, GELU, compress back."""
  
      def __init__(self, cfg: GPTConfig):
          super().__init__()
          d_ff = 4 * cfg.n_embd            # the classic 4× expansion (slide: d_ff = 4d)
          self.fc1 = nn.Linear(cfg.n_embd, d_ff)   # W1: expand  (C → 4C)
          self.fc2 = nn.Linear(d_ff, cfg.n_embd)   # W2: compress (4C → C)
          self.dropout = nn.Dropout(dropout)
  
      def forward(self, x):
          # expand → GELU (GPT's choice over ReLU) → compress → dropout
          # � Ask: "Why is the non-linearity essential?" → without it, fc2(fc1(x))
          #    collapses into a single linear layer; depth would buy nothing.
          return self.dropout(self.fc2(F.gelu(self.fc1(x))))
  
  
  # �����������������������������������������������������������������������������
  # â‘¥ TRANSFORMER BLOCK                  [Slide: "One full transformer block"]
  # �����������������������������������������������������������������������������
  # 💬 "This is the repeating unit — the slide diagram in 3 lines of code.
  #    PRE-norm: LayerNorm goes BEFORE each sublayer (GPT-2/modern convention),
  #    and the residual 'x +' is the highway that lets gradients flow through
  #    deep stacks."
  # [Slide: "Residual + LayerNorm"]
  
  class TransformerBlock(nn.Module):
      """Pre-norm block: LN -> Attn -> +residual -> LN -> FFN -> +residual"""
  
      def __init__(self, cfg: GPTConfig):
          super().__init__()
          self.ln1 = nn.LayerNorm(cfg.n_embd)   # γ, β — learned (2 × n_embd params)
          self.attn = CausalSelfAttention(cfg)
          self.ln2 = nn.LayerNorm(cfg.n_embd)   # second LN, own γ, β
          self.ffn = FeedForward(cfg)
  
      def forward(self, x):
          # 💬 Read these aloud as: "x plus attention-of-normalized-x" —
          #    the residual means each sublayer only learns a CORRECTION to x.
          x = x + self.attn(self.ln1(x))   # sublayer 1: communicate (across tokens)
          x = x + self.ffn(self.ln2(x))    # sublayer 2: compute (per token)
          return x
  
  
  # �����������������������������������������������������������������������������
  # â‘¢ THE FULL GPT                          [Slide: "Recap: what we just built"]
  # �����������������������������������������������������������������������������
  # 💬 Teaching tip: show __init__ + forward() FIRST as the bird's-eye view —
  #    it mirrors the recap-slide pipeline line by line — then descend into
  #    attention. People hold details better once they've seen the skeleton.
  
  class GPT(nn.Module):
      """The whole model: embeddings + N blocks + final LN + LM head."""
  
      def __init__(self, cfg: GPTConfig):
          super().__init__()
          self.cfg = cfg
  
          # Token embedding table: (vocab_size × n_embd), learned lookup.
          # [Slide: "Embeddings" — "initialized randomly, updated during training"]
          self.tok_emb = nn.Embedding(cfg.vocab_size, cfg.n_embd)
  
          # LEARNED positional encoding (GPT-style), one vector per position.
          # [Slide: "Positional encoding" — the 'Learned' flavor, not sinusoidal]
          self.pos_emb = nn.Embedding(cfg.block_size, cfg.n_embd)
  
          self.drop = nn.Dropout(dropout)
  
          # The stack: n_layer identical-shaped blocks, each with its OWN weights.
          # [Slide: "Stacking layers"]
          self.blocks = nn.ModuleList([TransformerBlock(cfg) for _ in range(cfg.n_layer)])
  
          self.ln_f = nn.LayerNorm(cfg.n_embd)  # final LN before the head
  
          # LM head: project (B, T, C) back to (B, T, vocab) — a score per token.
          # [Slide: "Output logits"]
          self.head = nn.Linear(cfg.n_embd, cfg.vocab_size, bias=False)
  
          # 💬 WEIGHT TYING: the output head and the input embedding SHARE one
          #    matrix (used transposed in the matmul). Saves vocab_size × n_embd
          #    params and works well in practice — mentioned on the logits slide.
          # ⚠� This is why the parameter count printout is ~16K lower than the
          #    worked example on the slides (which counts the head separately).
          self.head.weight = self.tok_emb.weight
  
          # Initialize all weights small and Gaussian (std 0.02, the GPT-2 recipe).
          # � "Why not zeros?" → all-zero weights = all neurons identical = no
          #    symmetry breaking; nothing distinct to learn.
          self.apply(self._init_weights)
  
      def _init_weights(self, m):
          if isinstance(m, nn.Linear):
              nn.init.normal_(m.weight, mean=0.0, std=0.02)
              if m.bias is not None:
                  nn.init.zeros_(m.bias)
          elif isinstance(m, nn.Embedding):
              nn.init.normal_(m.weight, mean=0.0, std=0.02)
  
      def num_parameters(self) -> int:
          # 💬 Compare the printout with the slide's worked example (~3.25M).
          # [Slide: "Parameter count: worked example"]
          return sum(p.numel() for p in self.parameters() if p.requires_grad)
  
      def forward(self, idx, targets=None):
          """
          idx: (B, T) token IDs
          targets: (B, T) next-token IDs (for training); None for inference
          returns: logits (B, T, vocab_size), loss or None
  
          💬 "ONE forward pass serves both training and inference — same brain,
             different loop. targets=None is the only switch."
          [Slide: "Training vs. inference"]
          """
          B, T = idx.shape
          assert T <= self.cfg.block_size, f"sequence length {T} > block_size {self.cfg.block_size}"
  
          # ── The recap-slide pipeline, line by line ──
          tok = self.tok_emb(idx)                                       # (B, T, C)  token IDs → vectors
          pos = self.pos_emb(torch.arange(T, device=idx.device))        # (T, C)     position 0..T-1 → vectors
          x = self.drop(tok + pos)                                      # (B, T, C)  X = E + position
          # ⚠� tok is (B,T,C), pos is (T,C) — broadcasting adds the same position
          #    vectors to every sequence in the batch. Worth pausing on.
  
          for block in self.blocks:        # n_layer blocks, each refines x
              x = block(x)
          x = self.ln_f(x)                 # final LayerNorm  [Slide: "Stacking layers"]
          logits = self.head(x)                                         # (B, T, vocab)
  
          loss = None
          if targets is not None:
              # ── TRAINING branch ──        [Slides: "Cross-entropy loss"]
              # 💬 "The model predicts the next token at EVERY position in
              #    parallel — T predictions per sequence, not 1. That's why
              #    transformer training is so efficient."
              # Flatten (B, T, vocab) → (B·T, vocab) and (B, T) → (B·T,)
              # because F.cross_entropy wants (N, classes) and (N,) of true IDs.
              # ⚠� cross_entropy takes raw LOGITS — it applies softmax + -log(p_t)
              #    internally. Don't softmax twice (a classic live-coding bug).
              loss = F.cross_entropy(
                  logits.view(-1, logits.size(-1)),
                  targets.view(-1),
              )
          return logits, loss
  
      # �������������������������������������������������������������������������
      # ⑨ GENERATION                  [Slide: "Generation: from logits to text"]
      # �������������������������������������������������������������������������
      @torch.no_grad()   # inference: no gradients, no backprop — weights frozen
      def generate(self, idx, num_new_tokens: int, temperature: float = 1.0,
                   top_k: int | None = None) -> torch.Tensor:
          """Autoregressively generate num_new_tokens tokens.
  
          💬 "The slide's loop: last logits → softmax → sample → append → repeat.
             One token at a time. This is how GPT writes a sentence."
          """
          self.eval()  # switches dropout OFF [Slide: "Training vs. inference"]
          for _ in range(num_new_tokens):
              # If the running text exceeds block_size, keep only the last
              # block_size tokens — the model can't attend beyond its context.
              # 💬 "This IS the 'context window' people talk about in big LLMs."
              idx_cond = idx if idx.size(1) <= self.cfg.block_size else idx[:, -self.cfg.block_size:]
  
              logits, _ = self(idx_cond)   # full forward pass; loss is None here
  
              # Take only the LAST position's logits — the next-token prediction.
              # (Training used all T positions; inference uses just one.)
              # TEMPERATURE: divide logits before softmax.
              #   <1.0 sharpens (more confident/repetitive), >1.0 flattens (wilder).
              # ⚠� max(temperature, 1e-8) guards against divide-by-zero at temp=0.
              logits = logits[:, -1, :] / max(temperature, 1e-8)
  
              # TOP-K: keep only the k highest-scoring tokens, set the rest to
              # -inf (so softmax gives them probability 0). Stops the model from
              # ever sampling a wildly unlikely character.
              if top_k is not None and top_k > 0:
                  v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
                  logits[logits < v[:, [-1]]] = float("-inf")  # v[:, [-1]] = k-th best score
  
              probs = F.softmax(logits, dim=-1)                 # scores → probabilities
              next_id = torch.multinomial(probs, num_samples=1) # SAMPLE (not argmax) (B, 1)
              # � Ask: "What changes if we use argmax instead?" → deterministic,
              #    and typically loops/repeats. Sampling is where variety comes from.
              idx = torch.cat([idx, next_id], dim=1)            # append & loop
          return idx
  
  
  # �����������������������������������������������������������������������������
  # ⑦ DATA: making (input, target) pairs
  # �����������������������������������������������������������������������������
  # 💬 "Where do the 'answer keys' come from? The text itself. The target is the
  #    input shifted one character to the right — free labels, no human needed.
  #    This is what 'self-supervised' means."
  #
  #    data:   [T, h, e, _, c, a, t]
  #    x  =     [T, h, e, _, c, a]
  #    y  =        [h, e, _, c, a, t]   � y[i] is the 'next token' after x[i]
  
  def get_batch(data: torch.Tensor, block_size: int, batch_size: int,
                device: torch.device) -> tuple[torch.Tensor, torch.Tensor]:
      """Sample batch_size random windows of length block_size from data."""
      # Random start indices; -1 leaves room for the shifted target.
      ix = torch.randint(0, len(data) - block_size - 1, (batch_size,))
      x = torch.stack([data[i:i + block_size] for i in ix])          # inputs
      y = torch.stack([data[i + 1:i + 1 + block_size] for i in ix])  # same, shifted +1
      return x.to(device), y.to(device)
      # ⚠� Random windows ≠ epochs. We sample with replacement, so "one epoch"
      #    isn't well-defined here — we just count steps. Fine at this scale.
  
  
  # �����������������������������������������������������������������������������
  # ⑧ TRAIN COMMAND        [Slides: "Why we minimize it" → "Training"]
  # �����������������������������������������������������������������������������
  # 💬 The training slide's loop, in code:
  #    batch → forward → loss → backward (gradients) → optimizer step → repeat.
  #    Map each line of the loop below onto that diagram as you scroll.
  
  def train(args):
      device = torch.device("cpu")  # CPU-only, by design for the workshop
      torch.manual_seed(1337)       # fixed seed → everyone in the room gets the
                                    # same loss curve. All of us get same model.
  
      # ── 1) Load data: just a plain text file ──
      if not os.path.exists(args.data):
          sys.exit(f"Data file not found: {args.data}")
      with open(args.data, "r", encoding="utf-8") as f:
          text = f.read()
      print(f"Loaded {len(text):,} characters from {args.data}")
  
      # ── 2) Build tokenizer FROM the data ──
      # 💬 "vocab_size isn't chosen by us — it falls out of the data. For tiny
      #    Shakespeare it's 65: letters, digits, punctuation, newline."
      tokenizer = CharTokenizer.from_text(text)
      print(f"Vocab size: {tokenizer.vocab_size}")
  
      # ── 3) Encode the whole corpus ONCE; split 90/10 train/val ──
      # � Ask: "Why hold out a validation set?" → train loss can fall from
      #    memorization; val loss tells us if the model GENERALIZES. Watch the
      #    gap between the two columns in the printout.
      data = torch.tensor(tokenizer.encode(text), dtype=torch.long)
      n_train = int(0.9 * len(data))
      train_data = data[:n_train]
      val_data = data[n_train:]
      print(f"Train tokens: {len(train_data):,}   Val tokens: {len(val_data):,}")
  
      # ── 4) Build the model from the 5 numbers ──
      cfg = GPTConfig(vocab_size=tokenizer.vocab_size)
      model = GPT(cfg).to(device)
      # 💬 Pause on this printout and reconcile it with the parameter-count
      #    slides (~3.25M). Slight difference = weight tying (head not double-counted).
      print(f"Model parameters: {model.num_parameters():,}")
  
      # ── 5) Optimizer: AdamW ──   [Slide: "From gradients to weight updates"]
      # 💬 "AdamW = the slide's `w -= lr × gradient`, but with per-weight adaptive
      #    step sizes from running averages of past gradients. Used by GPT-2/3,
      #    LLaMA — and by us."
      # weight_decay gently pulls weights toward 0 — regularization.
      optimizer = torch.optim.AdamW(model.parameters(), lr=3e-4, weight_decay=0.1)
  
      # LR schedule: linear WARMUP (first 100 steps), then COSINE DECAY to min_lr.
      # 💬 "Warmup: start gentle while weights are random garbage. Cosine: take
      #    smaller steps as we converge — like slowing down when parallel parking."
      # � "What's lr at step 0? At step `warmup`? At step max_steps?" → trace it.
      def lr_at(step: int, max_steps: int, base_lr: float = 3e-4,
                warmup: int = 100, min_lr: float = 3e-5) -> float:
          if step < warmup:
              return base_lr * (step + 1) / warmup            # ramp 0 → base_lr
          progress = (step - warmup) / max(1, max_steps - warmup)
          progress = min(1.0, progress)
          return min_lr + 0.5 * (base_lr - min_lr) * (1 + math.cos(math.pi * progress))
  
      # ── 6) Training loop ──
      # Fixed sample prompts so the audience can WATCH the same prompt improve
      # from noise → words → Shakespeare-ish as training progresses.
      sample_prompts = ["To be, or not ", "For I am falser than vows made in"]
      log_path = f"{output_path}/loss_log.csv"      # for graphing the loss curve later
      log_file = open(log_path, "w", newline="")
      log_writer = csv.writer(log_file)
      log_writer.writerow(["step", "train_loss", "val_loss", "lr"])
  
      # Derived intervals: 10 evals, 5 sample dumps, 4 checkpoints per run,
      # regardless of --max-steps.
      eval_every = max(1, args.max_steps // 10)
      sample_every = max(1, args.max_steps // 5)
      ckpt_every = max(1, args.max_steps // 4)
  
      t0 = time.time()
      model.train()  # dropout ON
      for step in range(args.max_steps):
          # Set this step's learning rate (PyTorch optimizers read it from
          # param_groups; we overwrite it each step with our schedule).
          lr = lr_at(step, args.max_steps)
          for g in optimizer.param_groups:
              g["lr"] = lr
  
          # �� THE four lines that ARE training ��  [Slide: "Training"]
          xb, yb = get_batch(train_data, cfg.block_size, batch_size, device)
          _, loss = model(xb, yb)               # 1. forward → loss (one number)
  
          optimizer.zero_grad(set_to_none=True) # 2. clear last step's gradients
          # ⚠� Forgetting zero_grad is THE classic bug: PyTorch ACCUMULATES
          #    gradients by default, so they'd pile up across steps.
          loss.backward()                       # 3. backprop: chain rule, automatic
                                                #    [Slide: "How backprop computes gradients"]
          optimizer.step()                      # 4. nudge EVERY weight a tiny bit
  
          # ── Periodic eval + logging ──
          if step % eval_every == 0 or step == args.max_steps - 1:
              model.eval()                      # dropout OFF for a fair measurement
              with torch.no_grad():             # no gradient bookkeeping needed
                  xv, yv = get_batch(val_data, cfg.block_size, batch_size, device)
                  _, val_loss = model(xv, yv)
              elapsed = time.time() - t0
              # 💬 Narrate the first line: loss ≈ 4.17 ≈ ln(65) — the loss of a
              #    UNIFORM guess over 65 chars ("Model B" on the cross-entropy
              #    slide). Watching it fall below that = the model is learning.
              print(f"step {step:5d} | lr {lr:.2e} | train {loss.item():.4f} | "
                    f"val {val_loss.item():.4f} | {elapsed:.1f}s")
              log_writer.writerow([step, loss.item(), val_loss.item(), lr])
              log_file.flush()                  # so the CSV is graphable mid-run
              model.train()                     # back to training mode
  
          # ── Periodic samples: the workshop's "wow" moment ──
          # 💬 Early samples are gibberish; mid-run grows words and line breaks;
          #    late samples look like a drunk Shakespeare. Same prompts each time
          #    makes the progress visible.
          if step % sample_every == 0 and step > 0:
              model.eval()
              for p in sample_prompts:
                  ids = torch.tensor([tokenizer.encode(p)], dtype=torch.long, device=device)
                  out = model.generate(ids, num_new_tokens=80, temperature=0.8, top_k=20)
                  generated = tokenizer.decode(out[0].tolist())
                  print(f"   sample: {generated!r}")
              model.train()
  
          # ── Periodic checkpoints (resume / compare across training stages) ──
          if step > 0 and step % ckpt_every == 0:
              save_checkpoint(model, tokenizer, cfg, f"{output_path}/ckpt_step_{step}.pt")
  
      # Final checkpoint — this is what the generate command loads.
      save_checkpoint(model, tokenizer, cfg, f"{output_path}/final_checkpoint.pt")
      log_file.close()
      print(f"Done. Wrote final_checkpoint.pt and {log_path}.")
  
  
  def save_checkpoint(model: GPT, tokenizer: CharTokenizer, cfg: GPTConfig, path: str):
      # 💬 "A checkpoint must contain everything needed to rebuild the model:
      #    1. the weights, 2. the 5 numbers (shape), 3. the vocab (so token IDs
      #    decode to the same characters). Forget the vocab → garbage output."
      torch.save({
          "model_state": model.state_dict(),   # every learned tensor, by name
          "config": asdict(cfg),               # the 5 numbers
          "vocab": tokenizer.vocab,            # the character list
      }, path)
      print(f"  saved {path}")
  
  
  # �����������������������������������������������������������������������������
  # GENERATE COMMAND — checkpoint in, text out
  # �����������������������������������������������������������������������������
  # 💬 "Inference = rebuild the exact same model, load frozen weights, loop
  #    generate(). No loss, no gradients, no optimizer — compare the two columns
  #    of the 'Training vs. inference' slide."
  
  def generate(args):
      device = torch.device("cpu")
      if args.seed is not None:
          torch.manual_seed(args.seed)  # same seed + same prompt → same output
          # 💬 Good demo: run twice with --seed 42 (identical), then without (varies).
  
      if not os.path.exists(args.checkpoint):
          sys.exit(f"Checkpoint not found: {args.checkpoint}")
      # ⚠� weights_only=False because our checkpoint also carries config + vocab
      #    (not just tensors). Fine for our OWN files; for untrusted downloads
      #    you'd want weights_only=True (it restricts unpickling).
      ckpt = torch.load(args.checkpoint, map_location=device, weights_only=False)
  
      # Rebuild the exact architecture and tokenizer the checkpoint was saved with:
      cfg = GPTConfig(**ckpt["config"])        # the 5 numbers → same shapes
      tokenizer = CharTokenizer(ckpt["vocab"]) # same vocab → same ID↔char mapping
  
      model = GPT(cfg).to(device)
      model.load_state_dict(ckpt["model_state"])  # pour the learned weights back in
      model.eval()                                # inference mode: dropout off
  
      # prompt → IDs → generate → IDs → text. The full round trip from slide 1.
      ids = torch.tensor([tokenizer.encode(args.prompt)], dtype=torch.long, device=device)
      out = model.generate(
          ids,
          num_new_tokens=args.num_new_tokens,
          temperature=args.temperature,   # � live demo: try 0.2 vs 1.5 and compare
          top_k=args.top_k,
      )
      print(tokenizer.decode(out[0].tolist()))
  
  
  # �����������������������������������������������������������������������������
  # CLI — two subcommands, as promised on the "Hands-on: the plan" slide
  # �����������������������������������������������������������������������������
  
  def main():
      parser = argparse.ArgumentParser(description="Tiny GPT: train and generate")
      sub = parser.add_subparsers(dest="cmd", required=True)
  
      # train: only data + steps are CLI args; architecture lives in GPTConfig.
      p_train = sub.add_parser("train", help="Train the model from a text file")
      p_train.add_argument("--data", required=True, help="path to UTF-8 text file")
      p_train.add_argument("--max-steps", type=int, default=2000)
      p_train.set_defaults(func=train)
  
      # generate: checkpoint + prompt + the sampling knobs from the slides.
      p_gen = sub.add_parser("generate", help="Generate text from a checkpoint")
      p_gen.add_argument("--checkpoint", required=True)
      p_gen.add_argument("--prompt", required=True)
      p_gen.add_argument("--num-new-tokens", type=int, default=500)
      p_gen.add_argument("--temperature", type=float, default=0.8)
      p_gen.add_argument("--top-k", type=int, default=40)
      p_gen.add_argument("--seed", type=int, default=None)
      p_gen.set_defaults(func=generate)
  
      args = parser.parse_args()
      args.func(args)   # dispatch to train() or generate()
  
  
  if __name__ == "__main__":
      main()
  
  

  ---

  You can read this blog using RSS Feed. But if you are the person who loves getting emails, then you can join my readers by signing up.

  Subscribe

---

• सिर्फ春 Removing .html Extension from URLs Without .htaccess

  If you are trying to remove .html from your URLs then you will find that online answers mostly rely on adding a .htaccess file to root. However, it is possible to achieve the same outcome by transferring your files from filename.html to filename/index.html. So, here’s a short Bash script to achieve that.

  # The following code is licensed under GPL-3.0-or-Later.
  # The full text of the license can be read at the repo.
  # https://codeberg.org/SirfHaru/remove-dot-html
  
  #!/bin/bash
  
  find -name '*.html' ! -name 'index.html' ! -name '404.html' |
  while IFS= read file; do
    folder=${file%.*}
    mkdir $folder
    mv $file $folder/index.html
    echo "moved $file to $folder/index.html"
  done
  

  The one major downside to this script is that people who were earlier using your filename.html link will now be met with a 404 error.

---

June 12, 2026

• Thejeshgn GN Weekly Notes 24/2026

  I am at IITM/Chennai for Paradox. I have been here since 2022, since the first Paradox. The fest and the crowd have grown bigger. I came by car with friends A&S. We took Old Madras Road and stopped at Sriperumbudur for a quick visit to the Rajiv Gandhi memorial.

  

  • I explored a new Food Embedding Model called Epicure. What I like about it is the methodology used for building it. It is easy to understand and replicate. It can also work as a method for most heterogeneous networks. I blocked a bunch of domains at home so my TV or other devices can’t be used in mass AI Scraping setups.
  • From next week, I will plan to publish the weekly playlists. It could be either songs I have just discovered, the most listened-to songs, or a mix. I wrote a small Marimo script to list my most listened-to songs of the week from my ListenBrainz listens. It doesn’t need an API key. Anyone can use it.
  • I couldn’t play this week. I need to find more playmates. I continued with my regular 2 P.T sessions. This week, we added dead hangs to my routine. Also did my personal best sets: 70 KG lat pulls and 50 KG seated cable rows. My body weight is 60 KG.
  • I would love to meet more folks. Also, I have learned to ask and take pictures during such meetings. I love that we have pictures. I plan to take quite a few pictures this coming week. I hope to post some.
  • You can set up your own tiny satellite ground station to receive packets from the CubeSats. Maybe one day.

  ---

  You can read this blog using RSS Feed. But if you are the person who loves getting emails, then you can join my readers by signing up.

  Subscribe

---

• सिर्फ春 It's Fine to not Know the Varnamala

  As a Hindi speaker, if you have stayed on mainstream social media for long enough you would have probably came across a version of a video which looks like this: a person asks strangers to recite the ABCs of English, and once they are done with that they ask for the varnamala. The people are usually able to perform the former task but not the latter. The message at the end of the video was clear: Hindi speakers should be ashamed to not even know the basics of their native language while being excellent in a “foreign” language.

  Aside from the “foreign” part there are various reasons as to why ignorance of the varnamala is not that big of a deal.

  For one, there is no clear reason as to why people need to learn the varnamala. I don’t have a clear memory, but I don’t think I was ever asked to do a कखग ever in my life. My family and friends taught me a lot of words and then the kindergarten taught me how to write those words; there was absolutely no involvement of the sequence of those characters. I think a lot of other people had this experience too. Even when they were taught the varnamala they forgot it after some time because there is almost no use of it.

  Other than practical considerations, there is also the problem of creating a false parallel with English. English has 26 characters while Hindi has almost double of it. Hindi varnamala is not consistent like English. And most importantly, as mentioned above Hindi characters are taught in a piecemeal fashion unlike English where you are supposed to memorize the whole string of ABCs.

  So the fact that people are not able to recite their alphabets should absolutely not considered a shame.

---

June 11, 2026

• Thejeshgn GN Exploring Epicure the Food Embedding Model

  FlavorGraph is a large-scale graph network that combines data from over a million recipes with chemical compound information from 1,500+ flavor molecules to predict ingredient pairings. It uses graph embedding methods to represent foods as dense vectors, enabling data-driven food pairing suggestions that go beyond human or chef intuition. In FlavorGraph, the chemical and recipe context signals are fused at training time via a fixed metapath design, leaving no inference-time knob to adjust their relative weights in the final embeddings.

  One could call Epicure an enhanced FlavorGraph. It builds on FlavorGraph to produce 300-D embeddings, but instead of a single embedding that combines both chemical and recipe context signals. It has three embedding models Cooc, Chem, and Core. That way, as a user, you can choose the embedding you want. It also includes more recipes from other languages, not just English. Recipes from English, Chinese, Russian, Vietnamese, Spanish, Turkish, Indonesian, German, and Indian-English are included. They are machine-translated into English for this. It’s good to see Indian recipes covered as well.

  They have also normalized the raw ingredient strings to 1,790 canonical entries using an LLM. I found this interesting. I also found that this doesn’t necessarily cover everything that an Indian recipe needs. For example, there are only coriander and coriander_root. It’s usually coriander seeds or coriander leaves in India. In their list, there is no way to differentiate it.

  The most interesting part is how they traverse the graph to construct metapaths for training Metapath2Vec models. The three models follow different logic to construct it. But they all have the same architecture and hyperparameters: 300-dim embeddings, walks_per_node=100, walk_length=50, context_size=7, 5 negative samples, batch_size=32,768, lr=0.0025, 20 epochs, no warm restart.

  From the paper, the methodology (I have embedded the SVG flowchart below) is easy to understand and, if required, can be replicated. But the raw data is not available, though sources are mentioned. Maybe it can be replicated with a different dataset. They have also used LLMs quite a bit in the data and training pipeline. For me, constructing the metapaths was the most interesting part.

  1. Epicure-Cooc. Walks the Cooc graph: pure I–I random walks weighted by NPMI. No compound nodes.
  2. Epicure-Core. Walks the typed-compound graph and injects pure I–I walks at --ii_repeat=10 alongside the typed-compound metapaths. Edge transitions are weighted so I–C hops are not oversampled relative to the smaller I–I edge set. The resulting embedding blends chemical and recipe-context signal.
  3. Epicure-Chem. Walks the typed-compound graph but with –ii_repeat=0: the I–I templates are absent and the only walks the skip-gram sees are compound-mediated. The chemistry extreme of the family.

  

  You can explore the embedding online or using a simple script below. In my exploration, I found that it is okay to use the nearest neighbors for an ingredient in a recipe, in chemistry, or in mixed contexts. But I don’t think it’s at a level where we could just replace ingredients. It also doesn’t have any information about allergens, texture, etc. But it is small and can be used to build on top of it.

  

  # /// script
  # requires-python = ">=3.11"
  # dependencies = [
  #   "numpy",
  #   "huggingface_hub",
  # ]
  # ///
  
  from epicure import Epicure
  
  m = Epicure.from_pretrained("Kaikaku/epicure-core")
  
  print("neighbors : chicken")
  print(m.neighbors("chicken", k=5))
  
  print("neighbors : coriander")
  print(m.neighbors("coriander", k=5))
  
  print("slerp : rice, cuisine:South_Asian")
  print(m.slerp("rice", "cuisine:South_Asian", theta_deg=30, k=5))
  
  print("closest_mode : chocolate")
  print(m.closest_mode("chocolate", kind="factor", k=3))
  

  To run the above script, include the epicure.py file from the HF repo in the same folder. The epicure package on PyPI is a different package.

  # Output of above script
  neighbors : chicken
  [('pork', 0.5807677507400513), ('beef', 0.5712239742279053), ('chicken_broth', 0.5498523116111755), ('peanut', 0.5233361124992371), ('cream_of_chicken_soup', 0.5217206478118896)]
  neighbors : coriander
  [('cumin', 0.7044016122817993), ('scallion', 0.6711025238037109), ('chili_pepper', 0.6609357595443726), ('turmeric', 0.6468461155891418), ('chicken_broth', 0.6463753581047058)]
  slerp : rice, cuisine:South_Asian
  [('turmeric', 0.7607102225586673), ('mustard_seed', 0.756659547118539), ('fenugreek_seed', 0.7468295496269882), ('coriander', 0.7428115853809022), ('cumin', 0.7388300342030064)]
  closest_mode : chocolate
  [('F_15/M0', 'American sweet confections and dessert bases', 0.7751955986022949), ('F_7/M1', 'Sweet liqueurs and confections', 0.7553147673606873), ('F_8/M4', 'Sweet confections and dessert ingredients', 0.7396374344825745)]
  
  

  Definitions

  Word2Vec is a neural method for building word embeddings from raw text. Words appearing in similar contexts end up close together in the vector space. It comes in two variants skip-gram and CBOW.

  • skip-gram: predict the surrounding context words from a target word
  • CBOW: predict the target word from its surrounding context

  Metapath2Vec (PDF) It does random walks through the graph guided by a defined metapath, then feeds those walks to a skip-gram model to learn node embeddings. It works well for Heterogeneous graphs. Basically, it’s like creating sentences based on the metapath, then running Word2Vec skip-gram on them.

  • A metapath is a typed node sequence, such as author – paper – venue – paper – author. It’s predefined by manually by the model designer.

  ---

  You can read this blog using RSS Feed. But if you are the person who loves getting emails, then you can join my readers by signing up.

  Subscribe

---

• सिर्फ春 How I Prepared for CUET-PG Economics (I Really Don't Want to Write This)

  Disclaimer: I am making this post since I keep getting a lot of questions around this topic. For people who intend to give CUET-PG in future here’s a warning: this post will probably not be very useful for you and it might even increase your anxiety. Let me know if this post is missing anything.

  Initially I was not very worried about my admission into the masters course, that is until I read the question paper (PYQ) of JAM 2023. My performance wasn’t bad but it wasn’t spectacular either. My “concepts”, as coaching centers like to call them, were clear but I was losing a lot of marks due to calculations. I was now very scared.

  And so I started in January 2026, one month before the next JAM and two months before the next CUET-PG. I decided to just solve JAM PYQs as my preparation and call it a day at that. My usual daily routine looked like this: open laptop, solve a question and look for solutions online if I am not able to do them, repeat. I also used pomodoro technique and it worked wonders for me with my focus increasing substantially. I did this routine every time I wasn’t at the college. Then stopped the preparation in the last week before the exam. My performance hadn’t showed much improvement in terms of calculation. But I gave my test and was pretty confident that I would make it in to the top 50.

  Then after that I gave the CUET-PG without any preparation because of my high confidence. The week after the test the JAM results came up: I had only got a rank around top 100. I was depressed for a day, then I went back to my PYQ routine after I learned that IIT Delhi also has an internal exam for some extra seats. At this point I was very exhausted since all I was doing was look at PDFs full of questions all day long. I have now also started solving papers of some other popular tests.

  Thankfully, the misery was short-lived as I got my CUET-PG scores in April next month. I have got a good rank, good enough to join IGIDR and DSE which are often considered better than IITs for economics. I was happy. I immediately gave up all of my preparations for IIT Delhi’s internal exam. (I forgot to apply for ISI MSQE, so I didn’t even had the choice to prepare for that.)

  Anyway, I am very happy with my choice of IGIDR.

---

June 05, 2026

• Thejeshgn GN Weekly Notes 23/2026

  It’s been a good week in general. The work schedule is back under control. I am getting back to normalcy there. I also played after a long time on Sunday. We played singles again. It was tough but fun. It tests one’s stamina and technique more than doubles. I enjoyed it.

  I am back to physical P.T again. I have also started doing some stretching at home every day, thanks to Sandeep and friends. My goal for the year is to gain muscle and do some pull-ups. I measure lean mass and body fat with a yearly DEXA scan. My body fat was 30% last year. I am not aiming a lot. But a couple of percentage points reduction from last year is a win.

  And if I can dead hang for 2 minutes and do at least 3 proper pull-ups. I will consider it a win. Let’s see.

  

  • Added more albums from Kannada and Malayalam to MusicBrainz. My contributions to MusicBrainz are mostly driven by the music I listen to, so I can track them in ListenBrainz. I also updated Uma’s playlist on her radio. It’s no more baby rhymes now. It has carefully selected songs. Currently, U2026 has songs from Kannada, Malayalam, English, and Italian. I figured out a way to publish playlists and embed them in posts. Maybe from next week I can include a weekly playlist.
  • If you think you are a part of the majority, hate will never reach you. One day, it will reach you and your children, for sure.
  • Got confirmations from all NMG/2026 grantees that they have received the funds. NMG/2026 is officially complete. I have added a blog post about the micro-granting BoF discussions at FOSS/2026. If you have any questions or points to discuss, please add to the post or send me an email. If you think you are a part of the majority, hate will never reach you. One day, it will reach you and your children, for sure.
  • I like the Ways of Working series (on Substack) by Hiran. I had read about Viashnavi Murthy before and, for some reason, had not subscribed. I blame Substack. Thanks to Pradeep for bringing it to my notice again through Mehar’s story. I have subscribed now. I have sat on public benches and spent hours reading or waiting when I was a teen. Though some parts of the city still have public benches, they are vanishing faster. The world in general needs more public benches. Losing them is more than losing resting places, says Brune in this beautiful piece on Places Journal.

  ---

  You can read this blog using RSS Feed. But if you are the person who loves getting emails, then you can join my readers by signing up.

  Subscribe

---

• Thejeshgn GN Uma’s New Playlist

  We have updated Uma’s playlist on her radio. It’s no baby rhymes now. It has carefully selected songs. Currently, U2026 has songs from Kannada, Malayalam, English, and Italian. I have heard all of them and read lyrics as well to make sure it’s friendly. It is tough to filter out songs for toddlers, especially those with some kind of rhythm. Do send me your suggestions if you have any.

  On the technical side, I maintain the playlists on ListenBrainz and store the exported version in CouchDB as a public JSON document. It serves as the canonical URL. That way, in case I want to add additional attributes, either from mbzlists or my own in the future, it’s easy. It’s JSPF, not XSPF. Conversion between the formats is straightforward.

  I also wrote a simple web component (with Claude’s help) to embed the JSPF as a table in my blog. I could use it to publish weekly lists, too. I have embedded U2026 using the same below.

  ---

  You can read this blog using RSS Feed. But if you are the person who loves getting emails, then you can join my readers by signing up.

  Subscribe

---

June 04, 2026

• Thejeshgn GN Who Calls a Child Anti-National?

  There have been many incidents related to CBSE (and NEET). Nisarga has written a responsible, well-written tech disclosure blog post, followed by Sarthak on the breakage of process and rules. Bugs and breakages happen in large systems. But public exam systems need a much higher standard of transparency and accountability.

  Bugs shouldn’t be there; software/workflows should be tested well, and there should be a process. But even beyond that, incidents do happen, and all software/workflows have bugs or loopholes, sometimes obvious and sometimes not so obvious. The most responsible thing to do is to accept it, fix it, hold folks accountable, and continue with the work. Except for these, I think the system is doing everything else.

  While this was coming to light, there was a personal incident reported by a 17-year-old CBSE student, Vedant. It was a case where he tweeted about his misplaced answer sheets. The tweets got enough attention and soon attracted haters (it’s a strong word, but it’s true). Most notable amongst them is Ashok Shrivastav, host of Do Took on Doordarshan News, who tweeted (now deleted) that the young student was an anti-national. His Do Took has always been notoriously known for being anti-people.

  

  His tweet showed no empathy or intelligence. It was just pure hate for the kid and his ability to garner attention around the problem he was facing. And every report that came out supported his claim and put the system at fault.

  After the backlash, he deleted and tendered a non-apology to the kid and family. But then the harm was already done.

  

  Doordarshan is destined to become Radio Rwanda, especially with the people on it and people who encourage and fund it. If you want an example, you can try Do Took (Known for Anti-Opposition, Communal polarisation, misinformation, etc.). Twitter/X is a medium that just delivers that information virally. It’s been happening in the media and on Twitter for a while; I think a lot of people noticed it this time because CBSE concerns everyone, including the majority. Usually, targeted hate is aimed at minorities, dissenters, activists, etc. And this time, the hate was against the majority. Hence the massive outrage and non-apology. What the majority has to understand is that “They will come for you”.

  ---

  You can read this blog using RSS Feed. But if you are the person who loves getting emails, then you can join my readers by signing up.

  Subscribe

---

June 02, 2026

• aktsbot A guide to self-hosting web apps on a VPS

  A guide to self-hosting web apps on a VPS

  02 Jun 2026

  ---

  If you have not chosen a VPS provider, the following are what I've used

  • Digitalocean
  • Vultr
  • Hostinger

  So now you got yourself an account on a VPS provider and you plan to host your dream project(s) on it. Well done, fellow traveller. Let our journey commence!

  Table of contents

  1. Choosing a server
  2. Configuring our server
     1. tmux
     2. A new user
     3. Configuring ssh on the server
     4. Setting up a webapp on the server
        • DNS
        • Installing runtimes and dependencies
        • Application directories
        • Setting up nginx
        • Certbot HTTPS
  3. Finishing up

  Choosing a server

  Most VPS offerings start from a VM that has

  • 2 cpu cores
  • 25 GB ssd storage
  • 1 GB of RAM.

  This should roughly cost us around 5-6 USD a month.

  By all means this should be more than enough for most hobby projects. How much RAM and CPU are your projects consuming when they're running on your laptop/desktop? This should be the only deciding factor for choosing the specs of your VM.

  As for the operating system, if you are unsure, choose Debian GNU\Linux.

  Configuring our server

  When we provision a VPS, we get a public IP assigned to our VPS/server. To connect to the VPS we'd do something like this from our terminal

  dev$ ssh root@12.13.14.15 

  Digitalocean for example, enables root logins via ssh to our new server. For other VPS providers it maybe different. They may provide a normal user which can sudo to root.

  dev$ ssh admin@12.13.14.15 

  If you need to specify the private key file, the command would look like

  dev$ ssh admin@12.13.14.15 -i /path/to/private-key.pem 

  There's a handy config that let's you not worry about ips and usernames while connecting to a server. Create $HOME/.ssh/config if it does not exist and add the following at the end

  Host myvps Hostname 12.13.14.15 User admin IdentityFile /path/to/private-key.pem 

  And then you could do

  dev$ ssh myvps 

  and we are in our server. Please note that the .ssh folder and its content needs to have their permissions in order. If we screw up, run the following

  chmod 700 ~/.ssh chmod 600 ~/.ssh/id_* chmod 600 ~/.ssh/config chmod 644 ~/.ssh/*.pub 

  I usually keep all of my private and public keypairs in the $HOME/.ssh folder, so that it becomes easy for me to do backups.

  Lets login into the VM and get cracking

  server$ sudo apt update server$ sudo apt upgrade 

  This updates and upgrades the installed software in our server. Now, we'll install a few handy tools to make our server management life a bit easier.

  server$ sudo apt install tmux neovim git curl htop 

  1. tmux

  tmux should be the first thing you run when you login into a VPS. Why? In the event that our ssh connection dies, any commands we were running on the server would still be running inside tmux. Let's configure it a bit

  server$ nano ~/.tmux.conf 

  Put the following in it

  # Index starts from 1 set-option -g base-index 1 set-option -g pane-base-index 1 # Renumber windows when a window is closed set-option -g renumber-windows on # no login shell set -g default-command "${SHELL}" # 256-color terminal set -g default-terminal "tmux-256color" # use 256 colors instead of 16 # Add truecolor support (tmux info | grep Tc) set-option -ga terminal-overrides ",xterm-256color:Tc" # Mouse set-option -g mouse on # Reload ~/.tmux.conf bind-key R source-file ~/.tmux.conf \; display-message "Reloaded!" 

  Ctrl+o and Ctrl+x to save and quit out of nano. If you roll with vim, use it to edit the files mentioned in the guide instead of nano.

  Now run tmux

  server$ tmux 

  We should see a green bar at the bottom indicating that we are in a tmux session. Here's a quick video on how to use tmux.

  Whenever I login into my server, this is what I do

  server$ tmux ls 0: 1 windows (created Tue Jun 24 23:19:42 2025) server$ tmux a -t 0 # or which ever session 

  2. A new user

  If your VPS provider sets you up with the root account by default, then its a good idea to create a normal user.

  server# adduser jim 

  Fill in the prompts that follow. The add our new user to the sudo group so that we can run sudo commands. Replace jim with your preferred username.

  server# usermod -aG sudo jim 

  Let's check if jim has powers by switching to the account.

  server# su - jim 

  Let's do something only root can do, like copying over the .ssh folder in /root to jim's home folder so that jim can login via ssh.

  server$ sudo cp -r /root/.ssh ~/.ssh [sudo] password for jim: 

  Provide jim's password and the command should have succeeded. Fix permissions of the new folder

  server$ chmod 700 ~/.ssh server$ chmod 644 ~/.ssh/authorized_keys 

  Let's now try to connect to our VPS with this new user.

  dev$ ssh jim@12.13.14.15 

  All good?

  3. Configuring ssh on the server

  We need to make 3 changes

  • prevent root ssh logins
  • only allow ssh logins with private keys i.e no password logins
  • change port to something other than 22

  Have two terminals with ssh to our server open. And do the following in one of them.

  server$ sudo cp /etc/ssh/sshd_config /etc/ssh/sshd_config.orig server$ sudo vim /etc/ssh/sshd_config 

  Backups saves lives!

  Make the following changes in the file. Find the corresponding line and change it.

  Port 2202 PermitRootLogin no PasswordAuthentication no 

  If the line of interest has a # infront, remove it before saving. Otherwise, it'd be considered as a comment. For example

  #PermitRooLogin yes 

  should become

  PermitRootLogin no 

  Then save the file and exit. Restart the sshd service on the server

  server$ sudo service ssh restart 

  Logout from one of the terminals and connect to the server like so

  dev$ ssh jim@12.13.14.15 -i /path/to/private-key.pem -p 2202 

  If not able to connect, we have the other terminal from which we can debug. If all is well, make our lives easy by making the appropriate edits to $HOME/.ssh/config on our dev machine.

  Host myvps Hostname 12.13.14.15 Port 2202 User jim IdentityFile /path/to/private-key.pem 

  Check once again with

  dev$ ssh myvps 

  4. Setting up a webapp on the server

  It is hard naming things, so unfortunately the actual physical machine that runs our webapp and the software that it has to run to serve content - like our webapp, html files, images files etc is also called a web server.

  This guide will use a web server program called nginx. Almost all webserver software allows one to host multiple websites on a single vps/server. Apache pioneered this technique and this is exactly what we need to host multiple projects on a single VPS.

  Here's our hypothetical webapp

  • https://myapp.com is a react.js SPA.
  • https://myapp.com/api is a nodejs api that the above react frontend would consume.
  • https://api.myapp.com is the same nodejs api as above.

  Assuming, we bought the domain myapp.com ofcourse.

  4.1 DNS

  Head over to where you purchased the myapp.com domain(namecheap/godaddy/bigrock?) and add an A record with @ as sub-domain to it. Set the destination to our VPS's public IP.

  A @ 12.13.14.15 

  After this is done, all requests to myapp.com, foo.myapp.com, bar.myapp.com etc will all land on our new VPS.

  4.2 Installing runtimes and dependencies

  We are installing node.js here as our hypothetical webapp's backend needs it to run.

  server$ curl -fsSL https://deb.nodesource.com/setup_23.x -o nodesource_setup.sh server$ sudo -E bash nodesource_setup.sh server$ sudo apt-get install -y nodejs server$ node -v 

  Refer nodesource for latest instructions on how to install node if you are reading this in the very future.

  Now lets set up our database.

  a. sqlite

  If your application can make due with sqlite, then please go with sqlite. Then as your application grows, switch over to an database system like postgres or mongodb. Most ORMs used in backend codebases support switching database systems. So, a switch like this would not pose a challenge.

  server$ sudo apt install sqlite3 

  Put the following in ~/.sqliterc

  .headers on .mode column 

  Then to open a database

  server:/var/projects/dbs$ sqlite3 myapp.db 

  b. postgres

  Just a side note, its a good idea to run database systems like postgres, mariadb/mysql, mongodb etc on a separate VPS. More on that later.

  Onto installing postgres. Latest instructions for doing that are available on the postgres website. Run the following one by one on the server.

  sudo apt install curl ca-certificates sudo install -d /usr/share/postgresql-common/pgdg sudo curl -o /usr/share/postgresql-common/pgdg/apt.postgresql.org.asc --fail https://www.postgresql.org/media/keys/ACCC4CF8.asc . /etc/os-release sudo sh -c "echo 'deb [signed-by=/usr/share/postgresql-common/pgdg/apt.postgresql.org.asc] https://apt.postgresql.org/pub/repos/apt $VERSION_CODENAME-pgdg main' > /etc/apt/sources.list.d/pgdg.list" sudo apt update sudo apt -y install postgresql 

  Next we setup a database user with which we can access the application database from the myapp backend.

  server$ sudo su - postgres -c "createuser myappdbuser --pwprompt" 

  The above prompts you for a password for the myappdbuser.

  Create the application database, if not done so already.

  server$ sudo su - postgres -c "createdb myapp_db" 

  Now we grant acces to myappdbuser to myapp_db.

  server$ sudo -u postgres psql psql> GRANT ALL PRIVILEGES ON DATABASE myapp_db TO myappdbuser; 

  The connection string for your application would look like so

  postgres://myappdbuser:s3cr3tPass@localhost:5432/myapp_db 

  If you decided to host the database on a separate vps,

  server$ sudo cp /etc/postgresql/<version>/main/postgresql.conf /etc/postgresql/<version>/main/postgresql.conf.orig server$ sudo vim /etc/postgresql/<version>/main/postgresql.conf 

  Look for the line that says

  #listen_addresses = 'localhost' 

  and change it to

  listen_addresses = 'localhost,10.2.3.4' # the above is the private ip for the database vps 

  Restart postgres

  server$ sudo service postgresql restart 

  Now the connection string for your application would look like so

  postgres://myappdbuser:s3cr3tPass@10.2.3.4:5432/myapp_db 

  c. mysql/mariadb

  server$ sudo apt install mariadb-server server$ sudo mysql_secure_installation 

  In the prompts that follow

  Enter current password for root (enter for none): Switch to unix_socket authentication [Y/n] n Change the root password? [Y/n] Y 

  Set the new root password and we're all set. Lets create a new database user instead of using root for our day to day.

  server$ sudo mariadb 

  Next, we'll create the user

  MariaDB [(none)]> GRANT ALL ON announcing-litebb.md announcing-poof.md announcing-spb-lil.md autologin-sway-debian.md better-font-rendering-linux.md color-me-baby.md copy-paste-vim-slackware.md dwm-st-installation.md hello-lc230.md install-docker-linux.md javascript-callbacks-promises-async-await.md lubuntu-i3wm-awesomeness.md minimal-unix-setup-sowm.md multiple-ssh-keys-for-services.md no-more-blurry-fonts.md openbsd-getting-started.md polkit-thunar-mount.md self-hosting-webapps.md setting-up-neomutt-offlineimap-msmtp.md __skeleton.md__ slackware-install-lvm-luks.md slackware-kernel-patch-install.md slackware-post-install.md swaywm-on-debian-11.md vim-to-neovim.md xml.xml TO 'admin'@'localhost' IDENTIFIED BY 'p4ssw0rd' WITH GRANT OPTION; MariaDB [(none)]> FLUSH PRIVILEGES; 

  Please note, what we did here was making a user similar to root. We'll create an application specific user next that we'll use for our myapp webapp.

  MariaDB [(none)]> CREATE DATABASE myapp_db; MariaDB [(none)]> GRANT ALL ON myapp_db.* TO 'myappdbuser'@'localhost' IDENTIFIED BY 'N3wp4ssw0rd' WITH GRANT OPTION; MariaDB [(none)]> FLUSH PRIVILEGES; MariaDB [(none)]> exit 

  The connection string for our application would look like so

  mysql://myappdbuser:N3wp4ssw0rd@localhost:3306/myapp_db 

  If we decide to run mariadb on a separate VPS, make the following edit in /etc/mysql/my.cnf

  ... bind-address = 10.2.3.4,127.0.0.1 ... 

  Then restart the service

  server$ sudo service mariadb restart 

  The connection string will now look like so,

  mysql://myappdbuser:N3wp4ssw0rd@10.2.3.4:3306/myapp_db 

  4.3 Application directories

  The following is just a convention I use. Feel free to put your application whereever you like.

  server$ sudo mkdir -p /var/projects/myapp.com/{frontend,backend} server$ sudo chown -R $USER /var/projects/myapp.com 

  Now get the code into the folders. Let's start by building and copying over the frontend project from our dev machine.

  dev:frontend$ npm run build dev:frontend$ ls ./ ../ dist/ src/ package.json 

  The dist folder is what we want to deploy. Let's make an archive of it, so that we can copy it over easily.

  dev:frontend$ tar -czf dist.tar.gz -C dist/ . 

  Now lets copy the archive over to our server.

  dev:frontend$ scp -i ~/.ssh/private-key.pem \ -P 2202 \ dist.tar.gz jim@12.13.14.15:/var/projects/myapp/frontend 

  Let's extract the archive

  dev:frontend$ ssh myvps "cd /var/projects/myapp/frontend; tar xzf dist.tar.gz; rm dist.tar.gz; ls" 

  if all went well, we should see the built react project in that folder in our server. Here's a handy deploy.sh script that has all of the above. Put this in your project root and add to .gitignore

  #!/bin/sh npm run build tar -czf dist.tar.gz -C dist/ . scp -i ~/.ssh/private-key.pem \ -P 2202 \ dist.tar.gz jim@12.13.14.15:/var/projects/myapp/frontend # or # scp dist.tar.gz myvps:/var/projects/myapp/frontend ssh myvps "cd /var/projects/myapp/frontend; tar xzf dist.tar.gz; rm dist.tar.gz; ls" 

  Lets do the same for our backend code.

  dev:backend$ ls ./ ../ app/ .git/ node_modules/ package.json 

  Include all folders and files we need into the archive

  dev:backend$ tar -czf api.tar.gz node_modules app package.json 

  Now lets copy the archive over to our server.

  dev:backend$ scp api.tar.gz myvps:/var/projects/myapp/backend 

  Let's extract the archive

  dev:backend$ ssh myvps "cd /var/projects/myapp/backend; tar xzf api.tar.gz; rm api.tar.gz; ls" 

  Here's a handy deploy.sh for this as well

  #!/bin/sh # npm run build # - if its a typescript backend tar -czf api.tar.gz node_modules app package.json # or if its a typescript backend # tar -czf api.tar.gz node_modules dist package.json scp -i ~/.ssh/private-key.pem \ -P 2202 \ api.tar.gz jim@12.13.14.15:/var/projects/myapp/backend # or #scp api.tar.gz myvps:/var/projects/myapp/backend # NOTE: tweak this as per your project. ssh myvps "cd /var/projects/myapp/backend; tar xzf api.tar.gz; rm api.tar.gz; ls" 

  To run the backend, I use pm2. Create a file called pm2.config.cjs in /var/projects/myapp/backend with the following. You can check this in with git.

  module.exports = { apps: [ { name: "myapp-api", script: "./app/index.js", }, ], }; 

  And then I check if my project secrets in the file /var/projects/myapp/backend/.env are all set.

  Then to run the backend, its a simple

  server$ npx pm2 start pm2.config.cjs 

  To see the logs

  server$ npx pm2 logs myapp-api 

  4.4 Setting up nginx

  This is the webserver that we talked about earlier. Install it with

  server$ sudo apt install nginx 

  Now on to setting up a vhost files for our webapp.

  server$ sudo nvim /etc/nginx/sites-available/myapp_com.conf 

  Put the following in it

  server { listen 80; root /var/projects/myapp/frontend/dist; index index.html index.htm; server_name myapp.com; location / { default_type "text/html"; try_files $uri.html $uri $uri/ /index.html; } # assumes the backend is running on port 3000 location /api { proxy_pass http://localhost:3000; proxy_http_version 1.1; proxy_set_header Upgrade $http_upgrade; proxy_set_header Connection 'upgrade'; proxy_set_header Host $host; proxy_cache_bypass $http_upgrade; } } 

  Remember our hypothetical webapp's deployment plan?

  1. https://myapp.com is a react.js SPA.
  2. https://myapp.com/api is a nodejs api that the above react frontend would consume.
  3. https://api.myapp.com is the same nodejs api as above.

  1 and 2 are covered by the above nginx config. We need one more file for hosting our api in its own sub-domain.

  server$ sudo nvim /etc/nginx/sites-available/api_myapp_com.conf 

  Put the following in it

  server { listen 80; server_name api.myapp.com; # assumes the backend is running on port 3000 location / { proxy_pass http://localhost:3000; proxy_http_version 1.1; proxy_set_header Upgrade $http_upgrade; proxy_set_header Connection 'upgrade'; proxy_set_header Host $host; proxy_cache_bypass $http_upgrade; } } 

  Now to make nginx use these configs, we need to run the following

  server$ sudo ln -s /etc/nginx/sites-available/myapp_com.conf /etc/nginx/sites-enabled/myapp_com.conf server$ sudo ln -s /etc/nginx/sites-available/api_myapp_com.conf /etc/nginx/sites-enabled/api_myapp_com.conf 

  We can ofcourse create the vhost files directly inside the sites-enabled folder, but it wouldnt be best practise.

  Now we restart nginx

  server$ sudo service nginx restart 

  Anytime, we make a config change to nginx, check if there are any problems with the config with

  server$ sudo nginx -t 

  If all green, update nginx with

  server$ sudo service nginx reload 

  4.5 Certbot HTTPS

  By now, visiting http://myapp.com and http://api.myapp.com should have loaded our webapp and the webapp's api. But we need that sweet sweet https.

  server$ sudo apt install certbot python3-certbot-nginx server$ sudo certbot --nginx -d myapp.com -d api.myapp.com 

  certbot would ask a few questions and after that would give us the ssl certificates for our domains.

  Note: Let's Encrypt SSL certificates(the ones that certbot sets us up with) expires after 3 months. By installing certbot with apt, we get a systemd timer also installed for checking the certificates' expiry. We can find it with

  server$ sudo systemctl list-timers | grep certbot 

  Finishing up

  Always remember to keep you server patched with updates. A simple set of

  server$ sudo apt-get update server$ sudo apt upgrade server$ sudo apt dist-upgrade server$ sudo apt-cache clean 

  would be all thats needed to do this.

  Just a heads up, keep an eye on database software upgrades like postgres / mysql. apt will tell you the packages that are about to be upgraded when you run apt upgrade.

  P.S always backup your database before upgrading.

  And with that, well done webmaster!

  Happy Hacking & have a great day!

---

June 01, 2026

• Thejeshgn GN Micro-granting in India – BoF at IndiaFOSS/2026

  I have proposed BoF (Birds of Feather) session and discussion at IndiaFOSS/2026. The proposal is yet to be approved as of today. I am logging the same content here for readers of my blog post and also for record keeping. I will update in both places if there are any changes.

  Background

  I am Thejesh GN. In 2018, I started a yearly micro-grant of ₹1,00,000 in memory of my mother, called the Nagarathna Memorial Grant (NMG). The idea was to support meaningful work with no strings attached.

  The reason for the grant can be anything that matters to the applicant and the people around them. While I personally prefer Free and Open Source Software, Creative Commons, and public-interest work, that is not a requirement.

  Over the years, the size of the grant has grown from ₹1 lakh to several lakhs, thanks to friends and supporters joining in. In 2026, the grant reached its largest size yet ₹5.6 lakhs with many people contributing and supporting the process.The idea is also not to grow beyond ₹10 lakhs as of now. Personally, I would consider any grant handling more than ₹10 lakhs/year in 2026 is not a micro grant. Even with the growth, the core idea remains unchanged, small, direct, trust-based support for meaningful work.

  2027 will mark the 10th year of the grant. This BoF is an opportunity to discuss how we can encourage more micro-grants in India.

  IndiaFOSS 2026 is happening on September 26th and 27th at NIMHANS, Bengaluru. It is a Free and Open Source Software festival organized by the FOSS United community. There are student discounts and There also diversity scholarships available, so please do apply:

  Key Focus

  • Explore how micro-grants can enable meaningful creative, public-interest, community, and FOSS projects
  • Discuss how to build communities around micro-funding and peer-to-peer support
  • Understand the challenges involved and ways to overcome them
  • Encourage more people to start and run small independent grants

  Brief

  In India, most funding for creative, public-interest, and open projects comes through large institutional grants. While those are important, micro-grants ( small, decentralized, peer-to-peer gifts or funding ) can often be more accessible, flexible, and liberating for both the giver and the receiver.

  This BoF is an open discussion for FOSS contributors, creators, artists, writers, researchers, and community organizers to discuss about

  • the usefulness of micro-grants,
  • the realities of running them,
  • and how we can encourage more people in India to participate in or start their own micro-funding initiatives.

  Possible Discussion Points

  1. What is micro-funding?
  2. Who can fund, and who can receive?
  3. How small is “small”?
  4. What are the challenges?
  5. What makes it rewarding or fun?
  6. How do we encourage more people to participate?

  Discussion Format

  Round-table discussion.

  We will begin with a short introduction to micro-funding and why it matters, followed by an overview of how NMG has worked, lessons learned, and challenges faced over the years.

  The session will then open into a broader discussion and brainstorming on possible discussion points mentioned above.

  Moderators

  • Thejesh GN
  • Will update

  Note

  I will make small changes to this as I gather more feedback on the session. But the core remains the same.

  ---

  You can read this blog using RSS Feed. But if you are the person who loves getting emails, then you can join my readers by signing up.

  Subscribe

---

May 29, 2026

• Ravi Dwivedi Budapest Travel

  In September 2025, I attended the annual LibreOffice conference in Budapest, Hungary. This gave me an opportunity to explore the city, which I will cover in this post.

  Let’s start with the currency. Although Hungary is a part of the European Union (EU), it doesn’t use the euro as its currency. Instead, it uses Hungarian forints (denoted by “Ft”). During my time in Hungary, 1 Indian rupee was equal to 4 Hungarian forints.

  After reaching the Budapest airport, I bought a 15-day public transport pass. The public transport counter is after you pass customs and immigration. The pass allows unlimited use of public transport in the city. I had to show my passport and pay 5950 Ft to get the pass. The pass had my passport number mentioned on it. The public transport passes can also be bought at any of the tram stations as well.

  

  

  

  Budapest is a union of two cities—Buda and Pest—lying on opposite sides of the Danube River. My hotel—Corvin Hotel—was on the Pest side.

  Budapest had good public transport. The buses, metros, and trams complemented each other. For example, the airport didn’t have metro or tram connectivity, but it was served by the bus. Most of the metro was on the Pest side, with only a couple of stations falling in Buda. However, both sides had an extensive network of trams.

  Furthermore, the information about the public transport was easily accessible. For instance, the map of tram stops inside the trams also included the bus routes one could get after alighting at those stops.

  From the airport, I took a bus followed by taking a metro on the M3 line to reach within walking distance of my hotel.

  

  During the conference I would take the tram to the conference venue. The trams were modern and fast. They also had a smiley face at the front, which gave them a friendly look. It seemed like the trams were happily doing their job. The city also had a good pedestrian infrastructure along with separate cycling tracks.

  

  Budapest’s tap water is officially safe to drink, which was mentioned on a sticker posted on the wall of the bathroom of my hotel room. So, I did not need to buy any water bottles while I was there.

  On the 6th of September, I went on a sightseeing tour of Budapest with my Dione. Our friend Attila, who was a local (from Hungary), joined us. We went to the central market from our hotel by metro.

  If you read my post on Vienna, I mentioned that the metro stations don’t have AFC gates but ticket validators instead. Budapest’s metro also has the same system. If you buy individual tickets, you need to validate them using the validators on the station before boarding the metro. If you are using a public transport pass like I was, then you do not need to validate, and you can board the metro directly.

  

  In 10-15 minutes, we reached the central market. Attila showed us around. I bought a fridge magnet and paprika powder as souvenirs. Paprika powder is a signature spice of Hungary. It is mainly available in two forms—one is sweet and the other being spicy. I wanted the spicy one, but I didn’t get that in that market. Therefore, I had to contend with buying the sweet version. The sweet version isn’t sweet though, it is just not spicy. After bringing that paprika powder home, it is mainly used for food coloring. I like it though and use it frequently in my omelets and other dishes.

  

  

  At some point, Atilla had to join the The Document Foundation (TDF) sightseeing group, so we parted ways at the central market. Dione and I continued our sightseeing and decided to start with visiting the Hungarian parliament, which is a tourist attraction. It was because we were on the Pest side and the parliament was also on the same side, while other tourist attractions were on the Buda side.

  So, Dione and I hopped on a tram and went to the parliament. We got off at a tram station just outside the parliament. The parliament is the icon of Budapest. The building has a gothic architecture and colored brown and white. One can buy tickets and take an inside tour. However, we didn’t have a lot of time, so we stayed outside the building.

  

  After spending some time outside the parliament building, we took a tram to the Chain Bridge. As I mentioned earlier, Budapest has two parts—Buda and Pest—separated by the Danube River. To go from one of the sides to the other requires crossing a bridge. Although Budapest has many bridges linking the two sides, the main one is the Chain Bridge.

  We walked on the chain bridge to get to the other side. The bridge gave a good view of the Danube River. It also had a statue of a lion. The Buda Castle (another major landmark of Budapest) was visible from the bridge.

  

  

  After reaching the other side of the bridge (the Buda side), we sat on a bench for some time and then planned on where to go next. We decided to go to Fisherman’s Bastion, which is another tourist attraction.

  We used the OSMAnd~ app to figure out which bus to take and hopped on one. Soon we reached Fisherman’s Bastion, where we found a flight of stairs that led upwards. Upon climbing the stairs, we got a panoramic view of the city. It also gave us a good view of the Hungarian parliament across the river. Going further upstairs, we found a statue of Stephen I of Hungary. He was the first king of Hungary, getting the crown in the year 1900.

  

  I found Fisherman’s Bastion to be the best tourist attraction in the city. As mentioned earlier, it offers a panoramic view of the city, which I liked. I liked the arhitecture and open space there. If you find yourself in Budapest, I would highly recommend that you visit Fisherman’s Bastion.

  

  

  Next, we went downstairs and returned to where the bus dropped us. From here on, we walked in random streets to see the residential and non-touristy side of Budapest. It was not so random as we walked towards Batthyány tér metro station. Upon reaching the metro station, we found a café where we stopped for a while for some coffee. After injecting some caffeine into our blood, we proceeded to find a place to have lunch.

  

  For lunch, we decided to go to Rákóczi tér metro station after reading on the internet about the food options there. Upon exiting the metro station, we found a market inside a building that had a lot of shops, but most of them were closed.

  After roaming around inside a bit, we found an Italian place open and decided to eat there. The name of this place was Matteos. We ordered an eggplant parmigiana, a lasagna artichoke, and a classic tiramisu. It wasn’t very tasty but filled us up for the day.

  

  Budapest has four metro lines, and we had been to three of them, so we decided to try the remaining line, which was the M1 line. It is the oldest line in the city and has a different vibe than the modern lines. This line was opened in 1896, one of the oldest subway systems in the world.

  The coaches were much smaller than the other metro lines, and the seating arrangement was something you would expect from a bus than a typical metro train. We rode all the way to the last stop, Mexikói út. Upon going outside, we found out there wasn’t much to do here.

  At this point, I checked the map and realized that Heroes’ Square is just a couple of metro stations away. Heroes’ Square is a tourist attraction in Budapest. It is located in Zuglóa and is a historically significant place in Budapest. It has a monument which features the Seven chieftains of the Magyars.

  

  Here, our unlimited public transport pass was handy because if it was paid per trip, we would think of the stop as a “wasted” one because we would have to buy a ticket again, but in this case we could just hop on again without any regrets.

  

  So we took the M1 line again and deboarded at Hősök tere station, followed by walking to the square. After roaming around for a while, we saw a trolleybus and decided to ride on that.

  

  

  A trolleybus is an electric bus that is powered by overhead electric cables. It is like a tram but runs on roads instead of tracks. We got down at Dózsa György út metro station. Then we took a metro to our hotel.

  Before going to the hotel, we went to a place to eat something. We had coffee and lángos. Lángos is a deep-fried Hungarian dish, which looks exactly like the Indian flatbread bhatura. I found it tasty, but since it was deep-fried, that was almost a given.

  

  The next day we went to Vienna—the capital of Austria—which I have already posted about. Check it out here.

  I had a good time in Budapest, and it is a beautiful city with good public transport and some amazing sites to visit. That’s it for now, and see you next time!

  Credits: Thanks Dione and Badri for proofreading.

---

May 15, 2026

• सिर्फ春 How Good is Pure Randomization for Indian Competitive Tests?

  I got a very good rank in <insert exam name> without studying anything. I just chose anything lol

  — Some Redditor

  This comment felt a very curious claim to me. See the exam that person is referring to, CUET-PG, is a pretty big one and if randomization is seriously that good then it would be pretty stupid to answer the questions yourself.

  So I decided to check the marks of 2025 batch to see how likely the Redditor’s claim would be. Turns out a university called Nehru Gram Bharati has leaked the marks of all candidates, at least for the Economics section. (Fuck privacy bro!) The first result for “nehru gram bharti cuet pg economics marks” was for a PDF on their website.

  It is, of course, difficult to do any data analysis on a PDF. I started looking for an azad software to do the PDF to spreadsheet conversion but didn’t get any. Eventually I was forced to use the Adobe Acrobat’s online tool which I think did the deed nicely.

  I used readxl and dplyr for importing the file. Cleaning the data after that required some time but in the end I had a nice numeric vector:

  # load the spreadsheet
  marks <- read_excel("~/Documents/R/CUET/marks.xlsx") |>
    pull(MARK_TOT)
  
  # clean bad entries and convert type 
  marks <- marks[! marks %in% c("MARK_TOT", NA)]
  marks <- type.convert(marks)
  

  The final CSV after tidying is available at marks.csv. A weird thing about that CSV is that it only contains 15,843 entries though there are supposed to be 18,379 students in the list. I have zero idea what happened to the rest.

  I then looked up some basic things:

  median(marks)
  # 45
  
  sd(marks)
  # 40
  

  I also plotted a graph just in case I was missing something.

  

  The Redditor’s claim seem quite plausible in the light of the above data, even if we account for the 50 marks rise in cut offs this year. However, I focused my attention back to the comment. It says “I just chose anything lol”. Hmm. I think I have been looking at the problem in the wrong way. Pure randomization after all doesn’t have anything to do with the distribution of marks within the rest of the students.

  In CUET-PG each wrong answer deducts 1 marks and each right answer gives 4 marks. Empty answers don’t change the score. Since the expected value of each question is positive (0.25), it would be in the best interest of the exam taker who is using pure randomization to answer all questions.

  Assuming our person of interest answered all questions, the probability that they would have a score which is at least as extreme as the one they got (which was 100 marks) would be:

  pbinom(35, 75, 0.25, lower.tail = FALSE) |> 
    format(scientific = FALSE)
  # "0.00001411975"
  

  That is very very unlikely.

  For conclusion, pure randomization only has a terrible expected value of 18.75. I don’t think anyone should submit to such chances if they have an alternative.

---

May 14, 2026

• सिर्फ春 I Love My New Laptop!

  I had been desperately waiting for my first laptop since the last month and now it’s finally here.

  It is the Motobook 60 Pro. Yes, you heard it right: it is a laptop by Motorola. At the time of writing this, the only other laptop model made by them was “Motobook 60”. Since Motorola is owned by Lenovo, the firmware design and the bad thermal management are very Lenovoish. The outer blue design gives it a modern look and I was genuinely surprised when I found out that the boot button was on the side next to the USB Port.

  

  Of course, the first task at hand was to install Linux. However, I was not able to boot either CachyOS or Fedora correctly. In my frustration I even accidentally deleted Windows though I had some intentions of dual booting before that. I looked up hundreds of forum posts. Ultimately the problem turned out to be the thumb drive itself. This new fucker was corrupting all of the files and it was only after using another old USB stick was I able to make any progress.

  I quickly created a bootable thumb drive with CachyOS, Limine, and Hyprland. Suffice it to say that the setup didn’t lasted long as I wasn’t able to connect to my WiFi. I wasn’t even able to fix the font size in end-4’s Hyprland dotfiles.

  I reinserted my thumb drive and this time I chose KDE instead of Hyprland. It was nice, and I would have continued for longer if it wasn’t for me getting annoyed by the extremely long compiling process of R packages. Arch, unlike Fedora or Debian, does not have binary files for R packages, which means the user has to compile it themself. I would not recommend that shit even to my enemies. I tried to suffer silently for a bit longer but my OS soon became unbootable when I changed the background image of Limine but forgot to update the hash.

  Damn it! I have had enough. I installed Fedora KDE Plasma because of its pure black theme. A little modification here, a little change there, and I had a very nice and stable setup.

  

  After a few hours I have downloaded almost everything except WINE for emulating Microsoft Excel (which might be useful for following professors using it in their classes). I hope that that would not be a big pain in the ass.

---

April 28, 2026

• Abhijith PA Patience could've saved me time.

  If I had been patient, it would have saved me time. One such instance is following.

  From my early blogs, you might know I am using mutt to do email. Just after I get along with mutt, I started using notmuch. Because limit search in mutt is always a pain when you have multiple folders. And what better tool out there than notmuch-mutt to bind both these.

  notmuch-mutt provide three macros by default.

  macro index <F8> \
  "<enter-command>set my_old_pipe_decode=\$pipe_decode my_old_wait_key=\$wait_key nopipe_decode nowait_key<enter>\
  <shell-escape>notmuch-mutt -r --prompt search<enter>\
  <change-folder-readonly>`echo ${XDG_CACHE_HOME:-$HOME/.cache}/notmuch/mutt/results`<enter>\
  <enter-command>set pipe_decode=\$my_old_pipe_decode wait_key=\$my_old_wait_key<enter>" \
        "notmuch: search mail"
  macro index <F9> \
  "<enter-command>set my_old_pipe_decode=\$pipe_decode my_old_wait_key=\$wait_key nopipe_decode nowait_key<enter>\
  <pipe-message>notmuch-mutt -r thread<enter>\
  <change-folder-readonly>`echo ${XDG_CACHE_HOME:-$HOME/.cache}/notmuch/mutt/results`<enter>\
  <enter-command>set pipe_decode=\$my_old_pipe_decode wait_key=\$my_old_wait_key<enter>" \
        "notmuch: reconstruct thread"
  macro index <F6> \
  "<enter-command>set my_old_pipe_decode=\$pipe_decode my_old_wait_key=\$wait_key nopipe_decode nowait_key<enter>\
  <pipe-message>notmuch-mutt tag -- -inbox<enter>\
  <enter-command>set pipe_decode=\$my_old_pipe_decode wait_key=\$my_old_wait_key<enter>" \
        "notmuch: remove message from inbox"
  

  One for search, one for reconstructing threads and one for manipulating tags, which I missed.

  Now my impatient part. I have already mapped f6 for my folder movements and in my initial days of notmuch, I only use just search. So I never cared about the f6 macro provided by notmuch-mutt. As time goes by I got very comfortable with notmuch. I was stretching my notmuch legs. I started to live more on notmuch search results date:today tag:unread than more on the mutt index. To the problem, since notmuch-mutt dump all results to a temp maildir location, can’t perform flag changes back to the original maildir which was annoying, because we need to distinguish what mail you read and what not when you subscribed to most of all debian mailing list.

  I was under the impression that, the notmuch-mutt is not capable of doing so and I just went like that without checking docs. I started doing all crazy hack to sync these maildirs.

  I even started reading notmuch-mutt codebase.

  Later, I settled on notmuch-vim. Cause I can manipulate flags sync back from notmuch to maildir.

  And while searching for something, I accidentally revisited the the the notmuch-mutt macro page and saw the tag manipulation. I was like :( .

  If I read about the third macro patiently when added that to config, I could’ve saved time by not doing ugly hacks around it.

  I think I learned my lesson.

---

• Ravi Dwivedi A day in Vienna

  On the 7th of September 2025, my friend Dione and I had a day trip to Vienna—the capital of Austria. We were attending a conference in Budapest, Hungary, which is 250 km from Vienna. So, it was a good opportunity to visit Vienna.

  We took a morning train from Budapest to Vienna and got back to Budapest by night. However, booking these tickets turned out to be a bit complicated. There were many websites to book the train ticket—Hungarian Railways, Austrian Railways, and third-party sites such as Omio. All these websites had different prices for the same ticket.

  I booked the tickets from the Hungarian Railways website as it was the cheapest. The train from Budapest to Vienna was €13, operated by Eurocity. Also, I had to pay €2 for the seat reservation on top. The train from Vienna to Budapest—operated by Railjet—was €21, along with €2 extra for reservation again—making it €23. The tickets for the two-way journey added up to €38.

  The cost of these tickets varied depending on when one purchses them: the sooner you purchase, the lower the price. I bought my tickets 15 days ahead of the date of journey and paid just €38. In contrast, Dione booked just one day before her trip and paid around €100 for her tickets.

  As for the seat reservation, long-distance trains in Europe usually require paying extra for the seat reservation. This ensures that you get your preferred seat, such as a window seat or an aisle seat. Nevertheless, you will get a seat on long-distance trains because they do not sell more tickets than there are seats. Therefore, you will get a seat without reservation as well. However, we reserved our seats so that we can sit together. This helped us more in the return part of the journey—from Vienna to Budapest—which was more crowded than the train we took from Budapest to Vienna in the morning.

  On another note, reservation is mandatory on some trains in Europe, but ours wasn’t one of them. In addition, people also use rail passes, so an extra charge is required on top for reserving the seats for pass holders. On the other hand, local trains do not require seat reservations in general.

  Our train’s scheduled departure was at 08:55 from the Budapest Kelenfold station. We reached the train station 40 minutes before the train’s scheduled departure. The Kelenfold station had free Wi-Fi, which was handy because I didn’t have a local SIM.

  

  

  Our train arrived on time. I tried to find our coach number but could not find the numbers written anywhere on the side of the coach. Luckily, we were helped by a fellow passenger who directed me to look at the doors, where the numbers were mentioned clearly!

  Then we got into our compartment and took our respective seats. Our tickets were checked twice - once while the train was in Hungary and the other when in Austria. Showing the PDF of the train ticket on our mobile to the ticket inspector was good enough for the purpose. Austria and Hungary are a part of the open transit Schengen area, which means this was the extent of the border control checks we had to go through.

  

  The train also had free Wi-Fi, albeit with poor connection at times. There were no eatery options inside the train.

  We deboarded at the Wien Hauptbahnhof station in Vienna. The journey was 250 km and took 2.5 hours, reaching Vienna at 11:25, which was the scheduled time.

  

  

  

  At the station, we bought a 24-hour public transport pass from a ticket machine for €8. The pass includes unlimited access to all the public transport in Vienna for 24 hours. My pass was valid from the 7th of September at 11:34 to the 8th of September at 11:33. A single public transport ticket (from anywhere to anywhere) costs €2.4. A single ticket of €2.4 can be used once on any public transport in Vienna—trams, metros, and buses.

  Therefore, the pass is a good deal if you are going to take at least four public transport trips in a day. Unlike the public transport pass I got in Budapest, the pass in Vienna was anonymous and not tied to the rider’s name.

  

  We wanted to visit the Schönbrunn Palace. The palace was reachable by subway. In order to get to the subway station, we started by going outside the station. But it was not outside. So we came back inside the station building and realized that the subway was underground.

  We took the subway and deboarded at the Schönbrunn subway station—the closest one to the palace. The ride was smooth; the train was pretty silent.

  By the way, like Budapest, there were no AFC gates for boarding the subway in Vienna. The stations had ticket validators instead, where you are supposed to validate your tickets before getting into the subway.

  

  These validators are in place to ensure that you use your ticket only once. Unlike AFC gates, which are present in metros of most of the countries I have been to, the ticket validators don’t act as a physical barrier to enter the boarding area.

  If you board the metro without validating your ticket, you will be facing hefty fines upon getting caught. I have heard that the fine is around €100. On the other hand, if you have a public transport pass like we did, then you don’t need to validate it before boarding.

  In addition, there were no annoying security checks either, unlike in Indian cities. In the Delhi metro, for example, you would need to scan your bags and pass through a security check before getting to the AFC gates.

  

  Now back to the story, after alighting at the Schönbrunn subway station, we walked to the Schönbrunn Palace. One can roam around outside the palace and click pictures for free. To go inside, however, requires buying tickets. The tickets for the palace can be booked in advance on the internet. We didn’t take the tickets in advance, as we decided to visit the palace at the last moment.

  So we went to the ticket counter and found out that we needed to wait for 1 hour 40 minutes before going inside if we took the tickets at that moment. In addition, one ticket costs €44 (around 4000 Indian rupees). Since we had to return to Budapest in the evening and only had a few hours in the city, we decided not to go inside the palace. Instead, we clicked a few pictures outside the palace.

  

  The Schönbrunn Palace is a UNESCO World Heritage Site and is a historically significant place. It servedas one of the residences of the powerful Habsburg dynasty. The palace looked so good that my friend Dione said, “It seemed like the palace was built yesterday”. This remark applied to other parts of Vienna we went to. For example, the subway stations also seemed like they were built yesterday.

  

  Now, we wanted to go someplace to grab a bite. I asked my friend Urbec for suggestions on where to go. They suggested we visit the steps named Strudlhofstiege, which had the added benefit of being in a neighborhood with good bakeries and buildings.

  So, we took the subway and deboarded at the Roßauer Lände station, followed by walking around a kilometer to reach the stairs.

  

  

  

  On the way, we were also looking for a place to eat. Unfortunately, it was Sunday, and Vienna closes on Sunday. That means most of the shops—including bakeries and cafés—are closed. Only places like railway stations have shops open on Sundays.

  By the way, walking around in the streets of Vienna was a treat. The streets were not crowded (as it was not exactly a touristy neighborhood) and had good pedestrian infrastructure, with clean streets and separate cycling tracks. The buildings were also beautiful.

  

  

  After some walking, we found a restaurant open. I grabbed the menu to check the prices. A lady at the shop asked me what I was doing, and I told her that I was browsing the menu. She said that the menu was in German. I don’t know how she knew that we didn’t know German, but it seemed like a racist thing to be told.

  We roamed around further and found a café by the name of Blue Orange, where we ordered coffee and croissants. When we got our order, the waiter told us that they were having some issues, so they wouldn’t charge us for the croissant if it wasn’t good.

  

  My friend and I took a bite, and both of us didn’t like the croissant. After some time, the waiter came to us and asked whether the croissant was okay, to which we said no. Therefore, they didn’t charge us for the croissant. This was the first time something like this happened to me. It felt like I was in a different world. I added a small tip at the end for this gesture, which I had to put in a jar at the counter.

  The cappuccino I ordered was €4.50, while the espresso that Dione ordered was €3.60. The croissant would have been €3.60. I remember Paris having cheaper croissants!

  Then when the waiter brought our drinks out, they automatically gave me the espresso and Dione the cappuccino. Dione found this funny because there is a stereotype in her country (Australia) that men drink strong black coffee, and women drink milky drinks like cappuccinos. She found it interesting that this stereotype seems to exist in Austrian culture too.

  We hopped on a tram to reach the nearest subway station and went to the Wien Hauptbahnhof station to have something before we caught our return train to Budapest.

  

  At the station, I had Esterhazyschnitten and Punschkrapfen (thanks, Urbec, for the suggestion). The lady at the shop warned me that punschkrapfen had alcohol in it, to which I said okay.

  Esterhazyschnitten was a cake made of almonds, while punschkrapfen was a jam-filled sponge cake, soaked in rum. Esterhazyschnitten was my favorite out of the two. The punschkrapfen was too sweet for my taste.

  

  

  While the station was well-built, there were a couple of things about the Wien Hauptbahnhof station that we didn’t like. There were no seats inside the station, so we had to eat outside the building. Also, the toilets needed to be paid for. It costs 50 cents to use the toilets at this station.

  The Vienna train station had departure boards all over the place. So, we went to the platform our train was to arrive on.

  

  

  When our train arrived, we had some difficulty locating our compartment. This train was operated by a different company (Railjet) than the one we took in the morning (Eurocity) from Budapest to Vienna, and we were able to locate the coach numbers using the digital board at the station. Each compartment had a digital board next to it on the station displaying the coach number. However, that wasn’t the problem. Even after reading the coach numbers and trying to find ours, it didn’t appear where we expected in the sequence.

  When we were not able to find our coach for a while, we asked a ticket inspector of the train who was standing on the platform. He directed us towards the front side of the train. So we started running to the front side as we didn’t know how long the train stops.

  As we ran toward our coach, we found out that the engine of the back train was connected with the last compartment of the train at the front. At that point, we realized that the train was a combination of two trains. At a later station, the train on the back side parted ways and went towards Vienna Airport.

  

  

  We had a smooth journey and reached Budapest a couple of hours later.

  Vienna is a beautiful city; we enjoyed being there, and we would like to visit the city again!

  That’s it for now. Signing off. See you in the next one!

  Credits: Thanks to Dione and Badri for proofreading.

---

April 27, 2026

• Sahil Dhiman Weekly Notes

  Weekly notes is a genre where people chronicle their week on their blogs. Weekly notes are like a window. I love going through these, as they’re a steady stream of week on week happenings and progress in people’s lives. It shows people making efforts to improve: from basic things like learning to swim or drive, to planning long-term goals such as vacations, moving house, states, or even countries. In some cases, they carry internal monologues, thoughts, and anxieties. These are like a constant nudge for me to work on myself, like them.

  These are the weekly notes I read nowadays:

  • https://adhavansivaraj.xyz/tags/weekly-notes/
  • https://learningwala.in/tags/weekly-notes/
  • https://thejeshgn.com/category/weekly-notes/
  • https://shrayas.com/tagged/weeknotes
  • https://tracydurnell.com/category/weeknotes/
  • https://sathyabh.at/

  Most are there on Thejesh’s weekly notes planet which autoupdates when new posts arrive, usually starting on Friday evenings, and by Monday, almost everyone has posted.

  It reminds of a word from The Dictionary of Obscure Sorrows - Kenaway :

  the longing to see how other people live their lives when they’re not in public; wishing you could tune in to the raw feed of another human existence, in all its messiness and solitude—shimmying in place while brushing their teeth, squabbling over where to put the shoes, talking out their problems on solitary commutes—if only to give you something to compare your own life against, and figure out whether you’re bizarrely normal or normally bizarre.

  Close enough.

---

April 24, 2026

• सिर्फ春 Are Entertainers Useless?

  In a way, I agree with a lot of points in Nicholas Decker’s recent article Social Credit. Our society with its modern mass media has allowed a handful of people to dominate our attention, and these people aren’t usually the ones you would want to be on top of the attention economy. I think you might have plenty of people in your mind who fit that description.

  The core idea of Nicholas’ essay goes like this: sports, fashion, politics, and entertainment are kind of things which do not contribute much value to our society but still command our awe and, hence, we should discourage them in favour of scientists which actually do contribute to the progress of our society. In the words of Nicholas,

  It is not so bad if someone should go into sports because they are of little intelligence, and would amount to nothing if it were not for their body. But we should regard every person of intelligence, who could have done something which would actually improve the world, who goes into sports as a shameful tragedy.

  I am not very sure how they concluded that entertainment or musicians are a zero sum game since they haven’t provided any justification for that. They also included politicians in their criticism of “professions and industries whose existence is entirely zero sum”. This seems weird to me, especially considering the fact that Nicholas is from USA, a country which is currently not having a great time under its new set of poltical representatives. Nicholas doesn’t actually address how their criticism applies to politics, so the fact that they casually inserted that word seems wrong.

  However, ignoring those bits, I would point out that the essay relies on one central assumption; the assumption that scientists are much more smarter than the rest of the population. This assumption critically weakens the logic of the essay and makes the whole argument a lot less convincing.

  Scientists are not that smart when compared to the rest of us. They are able to do what they do because they have spent a lot of time on it, and in that way they are not that different from experts in other fields. I think that the image of scientists as people with some innate talent is derived from the image of science as a meritocracy where the best idea always win. This is of course not true. There is also plenty of evidence that more educated people are more polarised on important topics of contention such as climate change or universal basic income.

  This weakens Nicholas’ arguments, since if scientists are really not that different from the rest of us, then transferring people from fields like sports or fashion may in fact lower welfare as people who previously enjoyed their jobs will be forced to do things which are possibly less enjoyable. Maybe there would also be a decrease in the total amount of effort applied by the person. In fact, it could be better to transfer, say, care workers instead.

  I am also not sure if there is actually a scarcity of researchers right now. Maybe it’s just the case that research institutes are not getting enough resources to do stuff. In that case a transfer from sports will not be required. Unfortunately, Nicholas hasn’t discussed this either.

---

April 21, 2026

• Ravi Dwivedi LibreOffice Conference Budapest 2025

  In September 2025, I attended the LibreOffice Conference in Budapest, Hungary, on the 4th and the 5th, and a community meeting on the 3rd. Thanks to The Document Foundation (TDF) for sponsoring my travel and accommodation costs. The conference venue was Faculty of Informatics, Eötvös Loránd University (ELTE).

  The conference was planned to be held from the 4th to the 6th, but the program for the 6th of September had to be canceled due to the venue being unavailable because of a marathon in Budapest. So, all the talks got squeezed into just two days, making the schedule a bit hectic.

  The TDF had booked my room at the Corvin Hotel. It was a double bedroom with a window. The breakfast was included in the hotel booking. The hotel was walking distance from the conference venue. One could also take a tram from the hotel to reach the venue.

  

  

  3rd of September

  On the 3rd of September, we had a community meeting at the above-mentioned venue. I walked with my friend Dione to the venue. Upon reaching there, I noticed that the university had no boundaries and gates. This reminded me of the previous year’s conference venue in Luxembourg, which also had no boundaries or gates.

  In contrast, Indian universities and institutes typically have walls and gates serving as boundaries to separate them from the rest of the city. Many of these institutes also have security guards at the entrance, who may ask attendees to present proof of admission before allowing them inside. I was surprised to find that institutes in Europe, like the one where the conference was held, did not have such boundaries.

  The building where the conference was held was red, which happened to be the same color as the building for the previous year’s conference venue. I remember joking with Dione that the criteria for the conference venue might have been the color of the building.

  

  During the community meeting, we shared ideas on how to spread the word about LibreOffice. The meeting lasted for a couple of hours.

  After the community meeting, we went to the hotel for dinner sponsored by the TDF.

  

  

  4th of September

  On the first day of the conference, attendees were given swag bags containing a pad, sticky notes, a pen, a conference T-shirt, and a bottle.

  

  The talks started early in the morning with Eliane Domingos, Chairperson of TDF’s Board of Directors, giving the inauguration talk. As always, I found Italo Vignoli’s talk on the importance of document freedom interesting.

  During the snack break, I noticed that there were three types of milk available for coffee: cow’s milk, lactose-free milk, and almond milk. Almond milk is rare in India, but I have managed to get it, but I have never seen lactose-free milk in India.

  Since I run fundraisers in my projects, such as Prav, I could relate to Lothar K. Becker’s talk. He discussed the issue that certain implementations in LibreOffice require a budget that is too large for any single interested entity to fund independently. Furthermore, The Document Foundation (TDF) cannot legally receive funds from government entities. Therefore, there is no organization or entity to pool resources from all the interested entities to finance the implementation.

  

  Another talk was by the Austrian Armed Forces on their migration to LibreOffice. I wanted to know why they migrated, and I found out that they did it for their digital sovereignty, and not for saving on the license costs. Another point presented in the talk was that LibreOffice is available on all the operating systems, while the Microsoft Office suite is not that widely available. The migration was systematic and was performed over a few years. They started working on it in 2021, and the migration was finished recently. In addition, it also required training their staff in using LibreOffice.

  

  The lunch was inside the university canteen. We were provided lunch coupons by the TDF. I got a vegan coupon with 4000 Ft written on it, which meant I could take lunch for up to 4000 Hungarian forints.

  

  

  During the evening, it was my turn for the presentation. I was done with preparing my slides ten days before my talk. I also got my slides reviewed by friends.

  My talk was finished in 20 minutes, while I was given a 30-minute slot. This helped us catch up on the schedule. Furthermore, I made my talk interactive by asking questions and making sure that the audience was not asleep. During my talk, my friend Dione took my pictures with my camera.

  My talk was on how free software projects could give users a say in freedom to modify the software. I illustrated this using the Prav project that I am a part of.

  After the talks were over, we were treated to a conference dinner at Trofea Grill. It had a great selection of desserts, which helped me sample some Hungarian desserts. The sponge cake was especially good.

  

  5th of September

  The next day—the 5th of September—I went with Dione to the venue early in the morning, as her talk was the first one of the day. Her talk was titled Managing Tasks with Nextcloud Deck. Later that day, I also attended a talk on Collabora. At lunch, I found the egg white salad quite tasty.

  

  

  After the lunch break, we had the conference group photo. I had a Nikon camera, which we used to take the group photo. I requested a university student to take our group photo and also taught her how to operate the camera.

  

  By the evening, the conference ended, after which we went to a pub, which was again sponsored by TDF. I had beer, but that one really tasted bad, so I couldn’t finish it. The only vegetarian option was goat cheeseburger, which my friend Manish and I opted for. The burger tasted awful. Apparently, I don’t like goat cheese.

  The next day I went sightseeing with Dione in Budapest. Stay tuned for our adventures!

  Credits: Thanks to Dione and Richard for proofreading.

---

April 20, 2026

• सिर्फ春 Hate as Distortion of Love

  …I feel that everything is about love and that hate is a perverse way of expressing love. It’s not that they are two sides of the same coin.

  — Gustavo Santaolalla, in an interview with TechRadar

  [SPOILER ALERT! THIS POST DISCUSSES THE PLOT OF THE LAST OF US.]

  One of the central themes of The Last of Us Part 2 is the contrast between Ellie’s hate for Abby and her love for Joel. In the story, the hate emerges from Joel’s brutal murder at the hands of Abby. Due to its importance to the story’s narrative, this idea of hate coming out of love heavily defines the music of the game. In fact, a lot of music from Part 1 was actually repackaged and given much harsher tones in Part 2.

  I am fascinated by this evolution of love, and for me this contains an interesting way to view “evil”. Of course, much of this has been already discussed in popular culture in the form of “humanising the villain”. However, inspite of this, the general consensus views evil as being a fundamental property of some outliers and seeks to punish them. After all, the populace generally finds it difficult to imagine, say, a Hitler who loves dogs and children. (There has been some evidence that that statement is actually true.)

  In my view, this only serves to provide a scapegoat instead of solving the actual problem. A better solution, in my opinion, would be to actually understand this evolution and try to reverse it. It would be difficult, especially considering we are talking about empathising with people who are sometimes even referred to as demons. However, inspite of difficulties it is actually the right method to adopt.

  In the world of The Last of Us it would be hopeless to imagine helping Ellie since the environment is a dystopian anarchy anyway. However, in more calmer state of affairs maybe there’s a chance of redemption.

---

April 16, 2026

• Sahil Dhiman What is Life (to you)?

  It started with a thought: to understand people’s perspectives on life and its meaning. So I texted folks, “What is life (to you)?”. Each of the following list items (-) is a response from a different individual, mostly verbatim.

  - A lot

  - Everyone has a few universal basic qualities, and some special qualities. To me life is pursuit of exploring world based on those qualities and maturing those qualities as one goes on about exploring world/life with those qualities.
  Discovering and enhancing experiences as one goes through them.

  - life is endless suffering

  - my answer might change daily, but this is what I’ve noticed and feel recently. Life is a spectrum with two distinct ends: what we control and what we don’t. At birth, the spectrum is largely tilted toward control, but throughout our lives, it gradually shifts toward the other side. Ultimately, as we approach death, we lose all control over any aspect of our existence, reaching the other end of the spectrum.
  tho this isn’t universal, privilege plays a huge part in what you control tho i believe it holds true for the majority
  but yeah man, meaning and purpose are dynamic, it’s in their nature to change i can give you a different answer this evening itself xD

  - Funeral Monologue from Synecdoche, New York. https://www.youtube.com/watch?v=Z9PzSNy3xj0

  - Zindagi ek nadiya hai, Aur mujhe tairna nahi aata
  (translation - Life is a river, and I don’t know how to swim)
  On a more serious note, Life is what you make it out for yourself. The only established truth is that it will end. We can never know if there is something after or if there was something before. So try to live a life that you feel aspired by? But this question was beautifully answered by that book which you had about that dying professor
  (Me - He was talking about Tuesday’s with Morrie)

  - My answer is 42

  - One, it’s living on your own terms, you define everything for yourself, success, normal, whatever. You get to curate your version of it no matter the societal norms.
  It’s an accumulation of experiences - friends, parents, work, activities, doing shit loads. Sab try karo- travel, zumba, art, music, workout, sports, dil kara ye karna hai karlo. (translation - If your heart wants to do it, just do it.)
  Then I think relationships - all that you’ve nurtured, people forget maintaining people because of work. It takes efforts to keep people in your life, everyone that comes has a place in yours, how well thats stays is upto you. You also get to curate your people, who stays who don’t. Family toh hai hi (translation - family is there) but everyone else that comes along can make it pretty good.
  So I don’t want to be 50 and be like chalo ab kuch apne liye karte hai… (translation - Come on, now let’s do something for ourselves) Do whatever shit you want today. Not everything costs money, and if it does get thrifty
  But do keep healthy while doing all of that

  - Being alive so that my daughter can grow up and i can help raising her kids as well. Raising kids without mother is tough :P

  - Definitively, I feel like Life is a by product of proteins and energy working together. But in a more personal sense, Life is a dumb joke played onto us. It’s a rat race. But rats exists because of life and then it becomes a chicken-egg problem
  Honestly, I don’t give good answers to life questions. I’m generally the one asking
  Life can be like a box of chocolates, you don’t know what you’re gonna get untill you experience the chocolate(assuming the chocolates are heterogenous and contains a mix of everything)
  Camus once said, “Life is a revolt�, and one of his students added more spice to it like “Life is a revolt against the meaninglessness of existence"
  I kinda feel like Life is the pursuit of every person’s search for meaning

  - Imprisonment waiting for execution 😄
  I have one more thought while we are on the topic , game with pre defined starting position and predefined destination , path to reach is a maze

  - A phase where you can have a really good time or really bad one, usually the mix of both. A phase where you are prisoner to responsibility and materialistic wants.
  It’s a hell for you, where you try to create heaven for others. Being born was never your choice, but ending is always in your hands but you are a prisoner. You fear that leaving this world behind will destroy the heavens you created for others and they will be back to hell. But eventually everyone moves on watching the hourglass of their life.
  Once you are left with no desires or no one to create heavens for, you look arround yourself. You see everyone chasing something, everyone scared of their limitted life time sliping away yet you want it to end sooner.
  Doesn’t matter if it was all good till now, or all bad. The other half is waiting for redemption. If it was all good, it’s best time to die don’t wait for the bad to start. If was all bad, it’s still the best time to die what if it was the good one and more worst is waiting for you. We desire to be remembered, yet we want to free from this loop of suffering.
  Someone once said, life is a suffering, chose your sufferings.

  - Life to me is to live without regrets and live with freedom.
  Life is always unpredictable and this unpredictability makes it more interesting and worth it.

  - As of now, for the state of mind that I am in , I think for me life is about subtle struggle, subtle inconveniences and yet moving forward cause that’s all I know.
  I am not sure if any of this has any meaning, but sometimes I feel I was born of a purpose and that the universe has my back.
  For me it’s about raising my consciousness, understanding people to their depths, gaining moderate material success and helping people to some extend.
  I have tried to seek a grander meaning but I have failed.
  Life for me is what I make out it.
  In my times of great success i rarely think about life for I am busy enjoying it, whatever you may call that state of mind.

  - For me its the little things that you enjoy with YOUR people

  - Life to me is about living and loving, and doing it in a way that sustains. It’s the people who shape you, the work you get absorbed in, the quiet moments in between. There’s also the wanting, the drive to figure out what’s worth going after and how to get there, but that’s just one part of it, not the point of it. And none of it happens in a vacuum. I’m aware of the privileges that let me live this way, and I try to hold on to that gratitude. In the end, life has both a material and a non-material side, and a lot of what we do is chasing material things in an attempt to satisfy something non-material within us

  - Mere liye (translation - for me) life is staying at my home and studying random economics papers. That’s when I enjoy myself the most.

  - Very complicated
  Some days I wish this life never ended and some time I feel it would be better if it stopped at that moment.
  It all depends on the events that happen in the so called “life”.
  So life to me is a string of events that happen anyway and you get to make some decisions which can turn it in any direction and then you wonder how did that happen.

  - not forgetting to breathe, learn, eat, game, take a good shit, love, sleep.

  - To be honest it changed with time! At 19 it was about freedom, wasn’t sure what freedom meant but i wanted that! To be free from everything, maybe because parents still controlled a part of my life. Then came 22-24 where i was working, trying to figure out what i want, the meaning changed from freedom to living for myself. To earn more, to be greedy about myself and pursue whatever would help me gain more steps in my career.
  Came my mba life, switched my life from doing for myself to trying everything out to have no regrets. Life meaning was just about living with no regrets, invested, gambled, did everything to earn that tag of “yeah, have tried that�. Now it has all switched to, it was all just a fake facade. Life turned to having a meaningful life rather than finding meaning in what i am doing. Living for people around me, chhoti chhoti cheezo m khushi (translation - happiness in small things(?)) isn’t really a topic of conversation but more of happy thing for me. So it changed, and m quite happy to be honest. Life did show me a lot of failures, but was privileged enough to face those failures. Gained a lot of learnings if not money😂
  Hopeful for more learnings and change meaning of life with time

  - A task.

  - You have different answers at different times You learn different meanings at different times When you are studying, basically it is about job, finding a partner then it becomes, house, car other things based on your income in between, there can be passion too
  Free Software was a passion, electoral politics too, but both kind of faded and I want cooperative and user driven development now (prav - something that motivates me every day) and these days learning Chinese and watching Cdrama takes a huge part of my leisure time it is heavily subjective and also influences by previous experiences people around you, how much influence they have on you
  it also depends on if they had to struggle in their life or not, for some life did not give much troubles and trouble itself can be relative people who never had to struggle may find even smallest challenges as troubles like if you own a car, your worry is finding a parking slot

  - I am too young to think about lyfe

  - A ticket to see the show on earth, I guess 😀
  I guess life is different depending on the mood. It is a very broad question.
  (Me - What is it in this present mood?)
  Learning stuff (like I am learning a new language) and being happy but also to regulate emotions in a world where being optimistic is getting harder each day.
  Life is also having a unique set of glasses you wear. Both in terms of looking from your eyeballs and your psychological perspective. Both are unique and cannot be replicated.
  It is interesting what people on their deathbed think of life. If I know I am dying, my perspective would change a whole lot.
  Life is finishing reading books while we are alive 😉
  Life is sleeping after a good XMPP chat 😉

  - Dukh dard peeda (translation - surrow pain suffering)

  - uhh to word it? life is just like a journey from A to somewhere and its all about what paths you take and what line you get on to me, just a series of short adventures that all connect to a larger sequence until you can’t have any more adventures-
  (Me - eee, THE END. drop dead, like a coin)
  yeaaaah- I am not really for spirituality of an afterlife, to me life just ends at some point, after which point there fails to remain a discernable you, and some X time after which, you will be last remembered, try to make that last time a good one I guess?
  (Me - no soul?)
  uhhh not in the way most people think of it i guess?
  theres just a lot of yous, theres the physical you, there is the idea of you, there is the expectation of you, and one of the undefinable you I would label as the soul maybe? like the part thats not physically you, but also certainly you
  (Me - can’t say I understood part, but I get you in this sense)
  mhm- well its about just questioning who you are more so questioning what life is-, I have sadly spent way too much time trying to figure that out

  - Making the best of the time you have

  - living a full range of experiences and embracing the good ones, seeing all that the world has to offer. In the end we were always just stardust. Might as well enjoy it when we are stardust with a consciousness of our own.

  - For some reason or the Universe’s /dev/random I was born here as a biological being, and from my experience I understood living is hard and the best way to live is by embracing it. Loving everyone and everything around you. Be happy and joyful until you naturally say good bye to this world.

  - Life is being fucked by everything and you just have to figure out and try to stick to the things worth being fucked for

  Note: Following was transcribed from a audio message.

  - There are five conditions to become a life to survive in the environment. I think there’s five conditions by the biological definitions and reproduction is one of the factor virus is not considered a life form because it cannot reproduce on its own but technically it’s kind of a life because it reproduces using the DNA ability this is the biological definition. Do you want a philosophical definition?
  My definition is kind of the same except that you get life experiences along with it as a human. Extra benefits is that you are not an NPC. All other organisms are NPCs. But humans can interpret the world and change it to their liking. That is life in the case of a human. But then many humans are mostly NPCs. But they still can change the life. Okay, fuck this. Where is this even going?
  A human is an exception in the case of life, because human is not an NPC. Human can interrupt the world, human can change it to its liking, which is why we are such a successful organism on this planet. That is life to me. That’s a human. But all of this is kind of meaningless, because the biological impurity of a human being still exists, so you still have the urges to reproduce, which kind of makes it like just another organism. But then, humans are yet to evolve to overcome that biological imperative.

  I’m grateful for all the replies, outlooks, and subsequent conversations I got to have after this question with everyone. After all, it was a deeply personal question. It does fit in nicely with my definition of life:
  “Life is all about experiences and all the transient relationships one gets to have with folks we meet on the way.”

---

April 14, 2026

• Ravi Dwivedi Hungary Visa

  The annual LibreOffice conference 2025 was held in Budapest, Hungary, from the 3rd to the 6th of September 2025. Thanks to the The Document Foundation (TDF) for sponsoring me to attend the conference.

  As Hungary is a part of the Schengen area, I needed a Schengen visa to attend the conference. In order to apply for a Schengen visa, one needs to get an appointment at VFS Global and submit all the required documents there, which are then forwarded to the embassy.

  I got an appointment for a Hungary visa at VFS Global in New Delhi for the 24th of July. There were many appointment slots available for the Hungary visa. One could easily get an appointment for the next day at the Delhi center. There were some technical problems on the VFS website, though, as I was unable to upload a scanned copy of my passport while booking the appointment. I got an error saying, “Unfortunately, you have exceeded the maximum upload limit.”

  The problem didn’t get fixed even after contacting the VFS helpline. They asked me to try in the Firefox browser and deleting all the cache, which I already did.

  So I created another account with a different email address and phone number, after which I was able to upload my passport and book an appointment. Other conference attendees from India also reported facing some technical issues on the VFS Hungary website.

  Anyway, I went to the VFS Hungary application center as per my appointment on the 24th of July. Going inside, I located the Hungary visa application counter. There were two applicants ahead of me.

  When it was my turn, the VFS staff warned me that my passport was damaged. The “damage” was on the bio-data page. All the details could be seen, but the lamination of the details page wore off a bit. They asked me to write an application to the Embassy of Hungary in New Delhi stating that I insist VFS to submit my application along with describing the “damage” on my passport.

  I got a bit worried about my application getting rejected due to the “damage.” But I decided to gamble my money on this one, as I didn’t have time (and energy) to apply for a new passport before this trip.

  Moreover, I had struck down a couple of fields in my visa application form which were not applicable to me, due to which the VFS staff asked me to fill out another visa application.

  After this, the application got submitted, and it was 11,000 INR (including the fee to book the appointment at VFS). Here is the list of documents I submitted:

  • My passport

  • Photocopy of my passport

  • Two photographs of myself

  • Duly filled visa application form

  • Return flight ticket reservations

  • Payslips for the last three months

  • Invitation letter from the conference organizer (in Hungarian)

  • Proof of hotel bookings during my stay in Hungary

  • Cover letter stating my itinerary

  • Income tax returns filed by me

  • Bank account statement, signed and sealed by the bank

  • Travel insurance valid for the period of the entire trip

  It took 2 hours for me to submit my visa application, even though there were only two applicants before me. This was by far the longest time to submit a Schengen visa application for me.

  Fast-forward to the 30th of July, and I received an email from the Embassy of Hungary asking me to submit an additional document - paid air ticket - for my application. I had only submitted dummy flight tickets, and they were enough for the Schengen visas I applied for until now. This was the first time a country was asking me to submit a confirmed flight ticket during the visa process.

  I consulted my travel agent on this, and they were fairly confident that I will get the visa if the embassy is asking me to submit confirmed flight tickets. So I asked the travel agent to book the flight tickets. These tickets were ₹78,000, and the airline was Emirates. Then, I sent the flight tickets to the embassy by email.

  The embassy sent the visa results on the 6th of August, which I received the next day.

  My visa had been approved! It took 14 days for me to get the Hungary visa after submitting the application.

  See you in the next one!

  Thanks to Badri for proofreading.

---

April 08, 2026

• सिर्फ春 Nicely Pleased With Openly Available Lectures

  I am currently preparing for IIT Delhi’s internal test for Msc Economics. An hour ago I came across this question in the 2023 question paper.

  

  It’s a simple question, but I was unable to do it because I keep doing some annoying calculation mistakes (which I didn’t notice at the time).

  I started thinking that there must be some deficiency in my understanding of joint probability distributions, so I looked up a particular video that has been stuck in my to-view list for a few months. The first slide was this

  

  LMAO. The video did show that it was just a calculation mistake that I was making. OpenCourseWare Movement for the win!

---

April 07, 2026

• सिर्फ春 Problems With EconHindu Utility Analysis

  Elections for the student council of Economics Department are currently going in my college. Out of all promotional images that are being flooded in college group chats one in particular caught my eye because of its poor economics, so here are some of the obvious issues I find with it.

  

  For one the axes aren’t even properly labelled! It only reminds me of this

  

  Due to this it is hard to understand what they mean by a continuous budget line. If it corresponds to candidates, then it should have been discrete. However, lets say that instead of comparing themself with other candidates they are comparing themself with the whole of India; in that case the size is large enough to warrant a continuous budget line.

  What about the indifference curve? I think they want to say that that is actually a social indifference curve. However, this begs the question how did they calculated it? As anyone aware of basic utilitarianism can tell you, it is impossible to aggregate individual indifference curves without making some sort of philosophical basis. However, the image hasn’t specified exactly what constitutes this basis. Without these assumptions clearly stated it seems that the analysis was manipulated to make it tangent to the budget line exactly at their preferred point and this considerably reduces the legitimacy of their analysis. I suspect a case of Arrow’s Impossibility Theorem.

  Another thing to note are the two axes. Why are there only 2, and not 3 or more? A decision as complex as deciding members of student council must take more variables into consideration. In my opinion, at the very least a 5 dimensional marginal analysis was necessary for this study.

  Finally, the caption actually contradicts the whole graph given above. As the budget constraint shows there are other people on the line too. So how is it possible that the opportunity cost is 0? Are they trying to imply that all other candidates are not even on the budget line, perhaps even at origin? If that’s really the case then it should be regarded as a clear violation of “no personal attacks” rule and strict action must be taken against them. I recommend solving the 5 variable optimisation version of this problem by hand as punishment.

---

April 06, 2026

• सिर्फ春 The Two Flavours of Atheism

  I recently came across the fact that, at least in recent times, there has been 2 major strands of Atheism.

  These are, in words of David Hoelscher, Scientifc Atheism and Humanistic Atheism. Now here are some of my own views of these cultures which have been heavily influenced by his article New Atheism, Worse Than You Think in CounterPunch.

  Scientific Atheism corresponds to a philosophy with near religious faith in power of science and a very intolerant view of religions. However, the view is not as unbiased as it seems at first glance. Science has often been used in modern times to advocate for an eradication of the “lower class”, with its most popular example residing in eugenics under Nazi dictatorship. Similarly, the New Atheism Movement has also been accused of Islamophobic and Sexist tendencies. The “Science”, in fact, can be used to prove almost anything that your heart desires. An almost blind faith in the superiority of science is, hence, very dangerous for the functioning of any society.

  Humanistic Atheism, on the other had, is more “humanistic” or tolerant of religions. Instead of simply rejecting religion because “there is no evidence a god exists” like Scientific Atheists would normally say, it actually engages with religious literature. It is also more conducive to discussions since the actual goal, after all, is to get good agreements on other contentious topics like climate change, monopolization of capital, poverty, and so on.

  My thinking has always been very similar to one propunded by Humanistic Atheists. However, it was only recently that I came to realize its deeper implications.

  Here are a few instances in my life of “unscientific” thinking: A believes in black magic, M believes in ghosts, T believes in Astrology, and RU believes in “cat crossing path is bad omen.” I usually have a very negative view of these stuff and on multiple occasions I have gone passive aggressive when talking to these people. But in light of the two distinctions that I just drew above regarding Atheism was this response on my part a sensible behaviour?

  I contend no. Rage over minor “unscientific” abnormality like these is very counterproductive. It angers people with no change in the actual status quo. It can also be considered very demeaning. A better idea is to always move to other issues which are far more deserving of attention. Humans deserve dignity, even when their opinions seem pretty stupid.

---

April 05, 2026

• सिर्फ春 Addresses are Really Helpful

  I was recently reading Demographic deposit, dividend and debt by Sonalde Desai. Following the Standard Operating Procedure, I looked up this Sonalde person and turns out she is a researcher at a research institute in Delhi called National Council of Applied Economic Research. The work of this organization felt interesting to me, so I had to pause my studies and see its location on OpenStreetMap.

  Nothing. Looks like the place hasn’t been mapped yet. I searched for 11, Indraprastha Estate since that was mentioned on their contact page and this time I did find an object. A person called “n’garh” had added the address back in July 2014 and I am so glad that I was quickly able to push a changeset (#180878526). Another win for address mapping!

---

April 01, 2026

• सिर्फ春 NOTICE ME SENPAI!

  Feeling a lack of love in your life? Then look no further, cause I am your perfect boyfriend material!

  This very large piece of meat will provide u with:

  • hundreds of dad jokes
  • thousands of romantic lines that I copied from manga 😘
  • lakhs of shittiest news from azad software folks
  • crores of “Hey! I mapped in that place ten decades ago!”
  • arabs of random economic topics that are not part of my syllabus
  • kharabs of “I like buses more than metro”
  • and so on… (My knowledge of Indian Numbering System has ended.)

  Think this is the product of your dreams?! Then send me that “be my senpai” right fucking now.

  Issued in public interest.

---

March 30, 2026

• Sahil Dhiman MiniDebConf Kanpur 2026

  MiniDebConf Kanpur 2026 was held on 14th and 15th March 2026 at the Indian Institute of Technology Kanpur.

  Having a Debian conference in the North was something many folks wanted. Ravi started the discussion (with IIT Kanpur folks) almost 7 months before the conference. Lots of folks from Debian India joined in organizing the conference, which was nice. All the meeting notes and discussions were posted on the Debian India mailing list, a first.

  Despite all the efforts, the conference start was delayed due to logistical issues. Things went fine post Day 1 lunch. We had two days of almost full schedules. disaster’s Decentralising Indian Communication was an interesting talk, diving into decentralised communication.

  IIT Kanpur is a huge campus with nice footpaths and greenery. We got the opportunity to explore their HPC at the Computer Center post conference.

  Work has already started for MiniDebCamp Kochi. More details can be found on the wiki.

  Working to make this conference happen was different with all the challenges involved, but overall, everyone was happy with the outcome.

  

  Group photo. Click to enlarge

---

March 29, 2026

• सिर्फ春 Paul Romer Likes Charter Cities

  A few days ago, I came to know that Paul Romer, of Romer Model fame, actually has a more or less active blog since 1997! This was a surprise, considering that I thought that the guy must be old enough to die. So, I looked up YouTube videos on him and found a TED Talk with the title “Why the World Needs Charter Cities”.

  “Charter Cities”? Huh. I have never heard of that term. However, A casual “research” soon gave me so much interesting information that my mind exploded.

  The term Charter Cities was first coined by Romer himself in the TED Talk I mentioned above. Charter Cities are supposed to solve the problem of poor quality of institutional structures in least developed and middle income countries by allowing high income countries to develop cities and then giving them near absolute power to regulate everything there in a manner similar to how British controlled Hong Kong, with a few caveats.

  The idea is fascinating, not least due to the massive amount of allegations of neocolonialism made against Romer’s idea. “Are Charter Cities Legitimate?” by Rahul Sagar discussed these problems in detail and concluded that it’s the practical issues, not moral ones, that make implementing Charter Cities difficult. I would recommend the interested reader to give it a read, though I do not fully agree with the essay.

  Aside from the discussion on these problems, the thing that fascinated me the most was that Charter Cities have already been implemented by some countries, though not exactly in the manner Romer approves.

  Prospera of Honduras was the first one. The city of around 2000 people is controlled by Honduras Prospera Incorporated, a company controlled by venture capitalists. The legal structure of the city is mostly defined by the corporation with the national government of Honduras having little control over it. Multiple attempts by the national government and judiciary has been made to destroy Honduras Prospera Incorporated, but some stakeholders from USA have lobbied enough to stop any repeals of its existence. Currently the corporation is suing the government for an amount about 1/3 of the country’s GDP. Now similar Charter City projects are being started in Africa.

  All of this sounds super dystopian to me and it keeps reminding me of the fictional Night City from Cyberpunk 2077. In this world of Neoliberalism, what hasn’t been tainted? I don’t know.

---