Hi, I'm Bhagwat.

I like building software and figuring out how things work. Lately that's mostly language models. I try things out, and sometimes write up what I find interesting.

2026
  1. Teaching a 1.2B model to solve chess puzzles with SFT

    A 1.2B model went from 0% to solving 45.8% of chess puzzles.

2024
  1. Steganography with Python

    Hide a text message inside an image by nudging pixel values one bit at a time.