About Me

My core expertise lies in building AI systems with production-grade reliability. I build LangGraph agents with tool use and checkpointed state, and retrieval grounded in real data — Postgres full-text search and Qdrant. At Melio the model never writes nutrition numbers: the server computes them from USDA data. I design structured output validation, a live cross-provider fallback (Anthropic ⇄ a self-hosted model), pre-release evals with RAGAS, DeepEval and MLflow, and observability with Langfuse so every token and cost is tracked.

On the data and backend side, I build data platforms and the APIs on top of them. At GovChime (~7M federal contracts, ~94M award transactions) my main work is the ETL: I moved SmartSync onto Temporal — 32 schedules on 6 task queues, in production since September 2026 — and it proves its completeness against SAM.gov's own files (0 of 80,922 active notices missing). PostgreSQL is the write model; CDC feeds a ClickHouse read model kept in checksum parity, and a logical-replication mirror keeps a second PostgreSQL in sync. On top sit an Express + Sequelize API and a Next.js frontend on Cloudflare Workers, where I cut a headline query from 13.6 s to ~1 s and tuned the caching layers. I also build event-driven architectures and background job processors with BullMQ and NestJS.

For frontends, I work with Next.js, React, and TypeScript to build fast, SEO-optimized applications. I care about performance, accessibility, and clean component architecture. Whether it's server-rendered pages for crawlability or real-time dashboards with WebSocket updates, I choose the right rendering strategy for the job.