Applied AI Engineer · LLM agents, RAG, evals & MCP · Fullstack TypeScript & Python
AI and fullstack engineer with 4+ years shipping production systems, now focused on LLM products that are safe to put in front of real users. I design systems where the model interprets and deterministic code decides: every LLM output is validated, tested and auditable. Applied in regulated healthcare and fintech.
Results that matter
- Production LLM agent on WhatsApp (healthcare): 95% field-level accuracy, every extracted field revalidated before persistence
- Safety-critical refusals gated by an eval harness of 120 adversarial scenarios; deterministic disambiguation cut LLM calls by 70%
- RAG pipeline in Python grounding enterprise agents in a centralized knowledge base, under audit and compliance controls
- Creator of nestjs-docfy, an open-source API toolkit with an MCP server for coding agents: ~10k npm downloads in four months
- Banking modules (Pix, TED, Onboarding) delivered 3 weeks ahead of schedule · TTFB -72% · bundle -67%
- 100% validation across 200+ sessions, promoted to Lead on an AI and search quality evaluation program (TELUS International)
M.Sc. candidate in Computer Science @ UFBA, advised by Prof. Dr. Frederico Araújo Durão.
- Zero-shot cross-domain recommendation: how geometric corrections of frozen text encoders transfer preferences across domains (Books → Music, Movies, Games) · code
- PULSAR: multi-agent system with BDI agents and a RAG explainer for early detection of clinical deterioration · code
