I build production AI that earns trust.

I lead production AI engineering at Atlax360 across BIFROST, ORVIAN, and Polaris, and I run the frontend platform behind the enterprise product surface. The work is document intelligence, workflow runtimes, and internal assistants, all of it language models treated as product infrastructure inside live financial operations.

Beyond the AI product work I build the platform around it: retrieval systems, agent runtimes, full-stack services, and the harnesses that let coding agents and human engineers work in the same codebase without it turning into noise.

My bias is toward retrieval quality, context assembly, eval loops, traces, and reliability. Most production AI does not break because the model is weak. It breaks because the context is messy, the orchestration is vague, the tools are poorly bounded, and nobody defined what good looks like.

I have worked in machine learning since 2015, starting when TensorFlow landed. The first chatbot I shipped failed basic intent boundaries: ask it about the weather and it tried to sell insurance. Intent boundaries and evaluation were missing, which is most of the job.

Before AI I spent years on full-stack systems for banks and retailers: BBVA, Bankinter, PagoFX, El Corte Inglés, Decathlon. Regulated environments where defects hit money movement and customer data, and release windows did not move.

My day-to-day stack is Python, FastAPI, Next.js, TypeScript, PostgreSQL with pgvector, Supabase, and Redis, with OpenAI and Anthropic APIs when they are the right tool rather than a reflex.