Research-aligned engineering
Systems & projects
Artifacts at the intersection of LLM reliability, observability, and production infrastructure—plus applied AI in safety-critical domains. Each project is something I deploy, measure, and iterate on, not a slide-deck-only idea.
Production-oriented observability for AI apps: OpenAI-compatible proxy, request tracing, usage and cost analytics, threshold alerts, and incident-style triage. Built for teams who need to answer what broke, for whom, and at what cost when models, guardrails, or traffic patterns change. FastAPI backend, React + Vite frontend, Apache 2.0.
Agentic web app for sinkhole risk in the Winter Park, Florida karst district—XGBoost susceptibility plus Gemini for explanation, monitoring, and alert drafting. A case study in high-stakes decision support where model outputs must be contextualized, monitored, and tied to real-world sensors—not just a one-off prediction.