Juan Cruz Fortunatti — Agentic AI Engineer

built MATEvals, a Python framework that replays cases drawn from 10,000+ production agent traces: deterministic tool-call assertions, LLM-as-judge checks, behavior-path diffs across versions, regression gates before rollout

build and operate multi-agent vendor workflows with Pydantic AI and LangGraph, including an agentic catalog editor for 50,000+ vendors: names, descriptions, prices, availability and translations, with per-market toggles each brand configures for itself. ~2,000 vendor-accepted changes a month

led, designed and shipped a conversation visualization and annotation system for bot conversations across seven brands, used by 275 internal users across brands and markets, feeding labeled production conversations back into evals

cut trace payload volume by 97% after diagnosing production failures from oversized traces and slow Langfuse writes