My work spans backend engineering and applied AI. I build contracts, evaluation surfaces, retrieval layers, and control gates so agent execution remains observable and resumable, while sensitive mutations require a human decision.
| 🔌 Agent-to-system contracts Versioned CLIs, REST and MCP interfaces, fail-closed operations, and machine-readable errors. |
🧪 Effect-based evaluation State and tool-trace oracles, environment isolation, contamination audits, and regression checks. |
| 🔎 Retrieval and context Persistent BM25 indexes, corpus invalidation, heading-aware chunks, and source line pointers. |
🧭 Governed orchestration Schema-validated state, idempotent steps, resumable workflows, bounded retries, and human approval gates. |
| 🧩 Plugin infrastructure Declarative manifests, closed-schema validation, build writes confined to the output directory, generated artifacts, and publication guards. |
|
| Auditable Multi-Agent Orchestration Authority, evidence, delegation, and proof of completion. |
Blind Differential Review Isolated evaluation roles, oracle auditing, and controlled text-removal tests. |
| Persistent Procedural Learning Evidence-based procedural memory with invalidation and bounded autonomy. |
How I Run Agents Stable invariants, iterative steering, domain tools, and task-local composition. |
Python Go TypeScript SQL FastAPI PostgreSQL MCP Docker Linux
AI Engineer at Hetosoft Sistemas · Pursuing a B.Sc. in Physics at the University of Campinas (Unicamp) · Campinas, Brazil