Bio

Staff engineer focused on production LLM systems.

I lead Data & AI at TaxDown and still work hands-on: architecture, code, reliability, evals, and production operations.

2M+ customer-facing LLM interactions handled in production
~1M automated support messages in 2026
~96% accuracy in human-reviewed production audits
~80% CSAT on automated customer support replies

TaxDown

Head of Data & AI and member of the Executive Committee. I set the technical direction for AI and data while building the core systems myself.

I built Adri, a multi-agent system that reads tax documents, extracts data, writes into tax returns, analyzes errors, applies edits, and routes risky cases to humans.

Production AI

My work covers LangGraph agents, RAG, structured outputs, tool calling, LangSmith tracing, golden sets, human-in-the-loop audits, guardrails, and provider fallback across OpenAI, Anthropic, and Gemini.

The systems run against real users, regulated data, payment flows, uploaded documents, and support conversations—not isolated demos.

Before AI

I worked at Expedia, Infinia, SpringCM, Channel IQ, and HERE/Nokia on data platforms, streaming pipelines, big data systems, cloud migrations, and distributed backend services.

Education: Telecommunications Engineering at Universidad Politécnica de Madrid and M.S. in Information Technology & Management at Illinois Institute of Technology.