I build and study ML systems, with a focus on inference, evaluation, and reliability.
| Project | What you can inspect |
|---|---|
| Paged Attention Engine | Paged KV cache, continuous batching, and matched-memory benchmarks. The README reports both wins and losses against different baselines. |
| testguard | Agent hooks that detect test weakening, with held-out checks for source hardcoding. |
| claim-check | Checks an agent's completion claims against executable verification evidence. |
| EV charging access | Reproducible geospatial analysis of charging access and equity in San Diego. |
| EV grid siting — Phase 1 | Data acquisition, grid-capacity joins, candidate sites, and routing; accessibility scoring and siting optimization are later work. |
| MAC8 | Supporting hardware work: an int8 multiply-accumulate design with formal verification and an RTL-to-GDS flow. |
Project READMEs describe the measured results, limitations, and reproduction steps. Older class exercises remain available in the repository history.
