Sudharsan BabuThe Notebook
Vol. I · Issue Sept 2026
01Index02Daily Log03Projects04Writing05Blog06About07Contact
Section 04 — Essays

Long-form,
quietly opinionated.

Essays are what happens when a log entry refuses to end. Published when they're ready, which is roughly once a month.

  • № 07
    Jul 2026
    ESSAY

    The unreasonable effectiveness of a smaller model

    Why I keep choosing 340M-parameter models for problems the internet insists need 70B ones. A quiet argument for taste over scale.

    9 min read
    Read →
  • № 06
    Jun 2026
    ESSAY

    Evals are the interface

    If you can't articulate what 'good' looks like as a metric, you don't have a product — you have a demo. On writing evals before you write the model.

    12 min read
    Read →
  • № 05
    May 2026
    NOTE

    On writing a README as if it were a letter

    Documentation is a moral act. A short piece on why every README should begin with 'Dear reader,' — at least in your head.

    6 min read
    Read →
  • № 04
    Apr 2026
    POSTMORTEM

    Six failed attempts at making an agent do a boring thing

    A postmortem of a project that took eight weeks to do what a Python script does in eight lines. Everything I learned about the current state of agents.

    14 min read
    Read →
Sudharsan Babu
AI Engineer · Working Notebook
A slow-moving record of experiments, models, essays and half-formed ideas. Published irregularly, edited rarely.
hello@sudharsan.dev
GitHubXLinkedIn
© 2026 — Set in DM Serif
Made with