Projects

A few things I've built or am currently working on.

Reinforcement learning environment and agent for the solo card game Scoundrel, trained with MaskablePPO and potential-based reward shaping.

Python Gymnasium

Fine-tune Qwen3-0.6B on personal git history to generate commit messages locally on Apple Silicon with MLX.

Python LoRA

Founding Engineer, Full-stack college application platform using Next.js and Supabase with various LLM integrations.

Next.js Supabase

Building tougher browser benchmarks and custom CUDA kernels for optimized LLM inference.

LLMs Python

Agentic AI orchestration system using MedGemma, with a staged reasoning chain to produce high-quality structured reports.

CUDA Python

Real-time eye-gaze estimation with OpenCV + MediaPipe; latency-aware smoothing and calibration UI.

Python OpenCV