Viraj Tipnis

I build reinforcement learning systems and agent infrastructure, mostly for finance and legal workflows, with a focus on async execution, evals, and long-running agent behavior.

Recently I've been contributing to OpenAI's Agents SDK, Codex, and related agent tooling, while also working on TycoonLE and TycoonBench, a JAX-based reinforcement learning environment and benchmark for planning, coordination, and multi-step decision-making.

I've also contributed to products such as marimo, Renderlet, and Void, spanning reactive Python notebooks, WebAssembly rendering, and local AI coding workflows.