I'm a GCP-native ML systems engineer. I build production ML platforms end-to-end on Google Cloud — Cloud Run inference, Vertex AI training, BigQuery warehouse, Firestore state, Cloud Build + Artifact Registry on the deploy path.
PARROT is the working proof — a multi-tenant agentic platform with a model registry, MCP tool catalog, persona-driven chat, and an OWASP-tested adversarial eval harness. Caveats are content. Every accuracy number wears its sample size and its bias label. No "AI-powered." No "revolutionary." Just the number and what's hiding inside it.