Thiago Polati

AI Engineer building LLM applications — agents, RAG, and tool-use systems — backed by 15+ years of senior backend engineering.

I help companies turn AI ambition into production-shaped systems. The interesting work in LLM apps today isn't the model call — it's everything around it: retrieval that works on real data, agents that don't drift, tool-use with proper guardrails, observability you can debug under pressure.

Where I work:

- Multi-agent orchestration — specialized subagents, durable file-based memory, gated execution protocols, human-in-the-loop approval gates (architected and run daily in my own working environment)

- Structured outputs over typed schemas with Pydantic and the OpenAI/Anthropic SDKs

- Retrieval and tool-use pipelines in Python and TypeScript/Node, on Postgres + Drizzle and FastAPI

- Designing evals and observability for non-deterministic systems — because "it worked once in a notebook" isn't a shipping criterion

Backend depth matters. 15 years of building production APIs, data models, and CI/CD discipline means I treat LLM apps as systems — not demos. I design for failure modes, idempotency, and what happens when the model is wrong.

Best fit:

- Teams building agentic products or RAG-heavy applications who want senior backend rigor, not a model-call wrapper

- Founders and CTOs who value pragmatism over hype, and want someone who'll push back when an LLM is the wrong tool

Fully remote, Madrid-based. I work in English, Italian, Portuguese, and Spanish.