Skill → evidence
Every skill below links to the project(s) that prove it. The portfolio is the evidence.
Hosted LLM products
Full-stack, deployed AI apps, not demos in a notebook.
RAG & source-grounded output
Ingest, chunk, embed, retrieve, and generate with citations.
Agentic workflows & tool calling
Planner + manual tool-calling loop with human approval.
LLM evals & regression testing
Deterministic checks + LLM-as-judge, scorecards, release gates.
Observability: latency & cost
Per-run latency and token-cost tracking on every call.
Secure secret handling
No secrets in code; .env.example everywhere; mock-first.
Structured output contracts
Strict JSON schemas validated before render and test.
Full-stack TypeScript / Next.js
App Router, API routes, Tailwind, Vercel deployment.