Trilingual ML/AI Engineer specializing in NLP, model evaluation, and production-grade classification systems.
Dual-layer Chinese-English web filter for 13 sensitive categories in international K–12 markets, using linear regression token-frequency screening followed by context-aware mmBERT with per-category sensitivity tuning. Achieved 30× speedup via RAPIDS/TensorFlow GPU acceleration.
Open-source Locality Sensitive Hashing library implementing hyperplane hashing with a modular pipeline for vectorization, signature generation, and JAX-parallelized similarity scoring. Achieves 15ms average query latency at 1M scale; pip-installable with CI/CD tooling (Ruff + UV).
