I build software you can test, trace, and trust.

A research platform where the rules stay put.

For my thesis, I needed an RL environment whose behavior I could reproduce, inspect, and compare. I built the simulator underneath it too: a Rust engine with Python bindings, batched stepping, legality masks, explicit observation and action contracts, plus replay tooling for when experiments drift.

Problem

Complex game rules and unstable experiments

System

Rust core → PyO3 → Python training API

Evidence

Documented contracts, replay, CI, and benchmarks

Result

Reusable infrastructure and a thesis graded 12/A