I built GenBI, a governed Text-to-SQL agent that turns natural-language questions into DuckDB SQL and Plotly charts. Across a 32-query benchmark, it achieved 93.8% execution accuracy at 80.6 ms p95 latency.
For GenBI, I engineered a ChromaDB retrieval system and a YAML semantic layer, then added SQLGlot guardrails for read-only execution, catalog validation, and row-level security. A self-healing execution loop and a 69-test Pytest suite support the system.
I also developed a zero-backend AI assistant and portfolio at ryandev.site, running a quantized Qwen3.5-2B model in the browser with WebGPU and a WebAssembly fallback. A dedicated web worker handles model loading and streaming so the React interface stays non-blocking.
For MSARR Technology, I built and deployed a responsive product and specification platform for microwave chambers and phased-array antenna test systems. I refactored its React application into route-driven views and added accessibility, SEO metadata, and conversion tracking.

