Nico Mastakas
AI Engineer building LLM agents, RAG and evals in production.
Product engineer, not a feature engineer, now with AI in the stack. My edge isn't calling the model; it's building and operating the whole system around it: agents, retrieval, evals, and the token economics to run it in production. Six-plus years shipping solo, end to end.
Selected work
07 projectsHybrid search that doubles recall
Lexical + vector ranking surfaces strong candidates that pure semantic search misses.
How I engineer with AI, not just prompt it
Agents, RAG, evals and fine-tuning I can prove, plus the orchestration and token economics to run them in production.
CI for prompts
Golden sets + a calibrated LLM judge fail the PR when prompt quality regresses.
On-device receipt OCR at 90% of cloud
Local OCR + a quantized LLM reach ~90% of frontier cloud vision, offline and at $0.
AI bill review with a human in the loop
Claude vision extracts and reconciles every line item before a human signs off.
Two-way invoice sync that survives the real world
Internal invoicing and QuickBooks Online kept in lockstep through duplicate, out-of-order and dropped webhooks.
A to-do app that only manages today
No backlog, no carry-over. Just Must Do vs Bonus, and 'do it now' under five minutes.