I design and build AI systems across the full stack: retrieval-augmented generation, multi-agent orchestration, and the LLM infrastructure that runs them, with evaluation and observability designed in from the start. Alongside engineering, I conduct scientific research in AI with a focus on multi-agent systems.
| Project | What it does | Core stack |
|---|---|---|
| supply-chain-agent-orchestrator | Multi-agent logistics disruption response with a durable human-approval gate, an MCP tool server, and a written LangGraph vs CrewAI benchmark | LangChain, LangGraph, CrewAI, OpenAI, MCP, SQLite |
| disclosure-rag | Retrieval over financial filings that cites the exact page and region for every claim, with self-correction before answering | PyTorch, Qdrant, FastAPI, Pydantic, Docker, Hybrid retrieval |
| llm-gateway-observability | A single gateway in front of LLM providers: semantic caching, guardrails, rate limits, cost and latency tracking | Langfuse, FastAPI, Pydantic, Redis, EKS, Terraform |
Each project ships with a test suite, CI, architecture decision records, and technical documentation.
AI and agents
Backend and data
Cloud and operations
Tooling and quality
