Two of the highest-leverage things you can build with LLMs today: retrieval that grounds answers in your own data, and agents that actually complete tasks. I build both — with the evaluation and guardrails that make them trustworthy.
Retrieval-Augmented Generation (RAG)
- Ground model output in your documents, policies, catalog, or knowledge base so answers cite real sources instead of hallucinating.
- Full pipeline: ingestion, chunking, embeddings, vector store (ChromaDB / pgvector), retrieval, and re-ranking.
- An evals harness (ragas or a custom suite) so retrieval and answer quality are measured against a test set — the piece most builds skip.
AI Agents
- Tool-using agents that plan and execute real workflows — research-and-draft, ops automation, data tasks.
- Guardrails and human-in-the-loop checkpoints for anything that writes to a system of record.
- Built on hands-on experience with LangChain and the agent ecosystem (including a merged fix to LangChain’s agent tool-input parser).
MCP (Model Context Protocol)
- I ship MCP servers that expose your tools and data sources to AI assistants through a clean, current interface — so your systems plug into the AI tools your team already uses.
- A differentiating, up-to-date capability: standardized tool access instead of one-off integrations.