Research Engineer
Atul Saxena
I build the environments models are trained in, and the tools that show what they learned.
- RL Systems
- Interpretability
- ML Infrastructure
Research Engineer
Atul Saxena
I build the environments models are trained in, and the tools that show what they learned.
Synthesizing verifiable reward signals outside maths and code
A C++ framework that automatically infers formal constraint schemas from ideal responses and executes constraint solvers plus generated unit tests, to produce verifiable rewards in law, biology, chemistry and clinical reasoning.
RLVR produces emergent chain-of-thought without supervised fine-tuning — but only where a deterministic verifier exists. That means maths and code. For law, medicine and the sciences, researchers hand-craft reward functions one domain at a time.
status: Research framework