A small, curated set of open-source projects. I keep this page intentionally focused on work that reflects how I think about reasoning, evaluation, and system design, rather than maintaining an exhaustive list.
A unified, minimal interface for accessing large language models via a registry-driven router and a declarative request language (ThinkerQL). The system provides schema-based contracts, secure credential handling, comprehensive testing, and real-time observability. The current implementation supports OpenAI and local Ollama models, and is designed to explore disciplined access patterns, routing, and evaluation in multi-model LLM systems.
Blazingly fast Sudoku solvers for 4×4, 6×6, and 9×9 grids implemented in pure C99 using Donald Knuth’s Algorithm X with the Dancing Links (DLX) technique. Each solver models Sudoku as an exact cover problem and is optimized for its grid size with zero runtime overhead. This project is used as a verifier and reasoning playground for studying constraint satisfaction, correctness, and evaluation in reinforcement learning and reasoning systems.
I maintain additional private repositories related to production systems, evaluation frameworks, and applied AI infrastructure. Some of these may be open-sourced selectively over time.