AI/LLM Engineer. I build multi-agent systems with adversarial evaluation, and I write about the engineering work in public.

What I'm working on

• Book Genesis — open-source multi-agent pipeline for long-form content production. 10+ books shipped through it in under 30 days. MIT, agent-agnostic. The interesting engineering is the evaluation layer (Genesis Score + MiroFish multi-persona reader simulation).

What I think about a lot

• Evaluation infrastructure for LLM applications. Generation is easy; knowing whether the output is good is the actual problem.
• Multi-agent orchestration with file-backed state. Durable > clever.
• Adversarial audit as a hard gate before scoring. Optimizing against a metric without an adversarial check is how you get score inflation.

Stack

Python, Bash. Multi-agent orchestration across Claude Code, Codex, Antigravity, Kimi, and any file-based agent runtime. Prompt engineering at scale, eval harness design, persona-based simulation, structured generation, durable file-backed state for agentic workflows.

Open to

AI/LLM engineering roles — remote or relocation — at companies that ship real LLM-backed systems and take evaluation seriously. Based in Brazil 🇧🇷, fluent in English and Portuguese. Available full-time.

Where to find me

• LinkedIn: felipeloboai
• X: @FelipeL72767971
• Email: blacksheeptec@gmail.com