Jeet Ganatra

Engineering@Amazon. Tinkering with various stages of building Language Models and diving deep into what makes them learn.

currently reading on-policy distillation of language models latest blog rl for language models, from first principles jun 6, 2026 · 36 min read every derivation kept in open source qwen3-nano-math-reasoner
hand-drawn ink illustration of a token factory: a Victorian-era apparatus with GPU-like card racks, attention switchboard, and a stream of tokens emerging as a constellation
hyderabad, india · github.com/jeetganatra · @jeetganatra1