About

I’m José Henrique, an AI engineer in Porto Alegre, Brazil, and a master’s student at UFRGS. My research asks whether system-level telemetry (eBPF, kernel events) can tell you what an AI agent actually did, and catch a compromised one that the transcript misses.

My background is production LLM systems (multi-agent architectures, RAG and Graph RAG, MCP tools) and the data infrastructure under them: Airflow pipelines, Elasticsearch and Neo4j at tens of millions of records.

I write here about what I build, what I measure, and what breaks. If you run agents unsupervised, I’d like to hear how you find out what they did: contact@josehenrique.dev, LinkedIn, GitHub.