I build LLM workflows that survive real traffic — and prove it with traces, before/after metrics, and public repos.
I'm Marco — AI engineer building production workflows for teams that need automation they can trust at 03:00. The brief varies: document pipelines, agent orchestration, internal copilots, back-office automation. The shape is always the same — too much manual work, not enough headcount, zero tolerance for hallucinated answers.
Stack on the right. Every system ships with traces (Langfuse), typed guardrails (Pydantic), and a CI/CD path that doesn't depend on my laptop. If it can't be re-run against last week's input and produce the same answer, it's not production.
Currently shipping one project per month through 2027 — public repos, real users, measured before/after. Open to advisory engagements and scoped builds across LLM ops, agent systems, and back-office automation.
LANGUAGES
FRAMEWORKS
INFRASTRUCTURE
MODELS
DATA & AI-OPS
TOOLS
Currently between projects — next ships [MONTH]
Reply in under 24h. First call is a 30-minute scope review — no deck, no pitch. You leave with a written assessment whether or not we work together.
