AIDevelopment
I help product teams put real LLM features into production, not demos that work on Twitter and break in front of users. That means careful prompt engineering, retrieval that actually retrieves, evals that catch regressions, and a UI that fails gracefully.
Typical engagements
- RAG over your docs, support tickets, or codebase
- Agent workflows for internal automation
- Production-grade chat & copilot UX
- Evaluation harnesses + prompt observability
Stack I lean on
- Anthropic Claude · OpenAI · open-weight
- LangGraph · custom orchestration
- pgvector · Turbopuffer · Pinecone
- Modal · Inngest · Trigger.dev