A framework for the automated derivation and parallel execution of finite difference solvers on a range of computer architectures.

Automatic code generator for training Reinforcement Learning policies

A tool to parse "Waters/Supremica" files and generate Python code from his inside supervisors.

LLM agents that generate, verify, and evolve Triton GPU kernels. Includes a reward-hack-resistant benchmarking harness with strict correctness verification and fresh-input evaluation. Achieves up to 174.7× over PyTorch eager and outperforms FlexAttention (1.48×) and SDPA (1.17×) on selected workloads.