I'm a full-stack machine learning engineer who likes solving real-world problems. Previously, I've worked at Modal, a serverless compute platform free of config files, Edlight, PBC, a startup building easy-to-use solutions for teachers, and Procurement Sciences, a startup building proposal writing tools for government contractors.
I love building open-source projects and participating in competitions I think are cool. Some highlights:
- A minimal LLM inference engine in Rust.
- RLMs in Rust using RustPython and gVisor.
- A UNIX-like Rust OS built for RISC-V, available on the web.
- An interactive Street Fighter 3 demo against an RL-trained LLM in a sandboxed game engine.
- Winner of the GPU Mode practice round.
- An ultrasound substructure localization system as the final project for SCU's ENGR 110 course and a POC of LLMs for pixel-level vision tasks.
- admirer, a VLM based on CLIP and GPT-2 built nearly two years before any major provider release.
- chexray, a chest x-ray diagnosis and report-writing system that outperformed contemporary SOTA models with 8% of the prior required training data.
In my free time, I'm a climber. I've redpointed up to 8a+ (5.13c) on lead and 8A (V11) in bouldering. I've also onsighted up to 7c (5.12d) and 7B (V8). I also enjoy learning about health, cooking/baking, and reading.