What problem does it solve? Moving an LLM app from a notebook or CLI prototype to a live production service involves many failure-prone steps: extracting a REST API, externalizing state, building a lean Docker image, handling secrets safely, and tuning the vector store. This Skill provides a verified 9-step promotion workflow so nothing critical is missed. ## Core Features & Use Cases - Prototype-to-service extraction: Convert notebooks or CLI loops into a FastAPI microservice with caller-supplied user_id/session_id identity and guardrails before the chain. - Production containerization: Multi-stage Docker builds with uv, non-root users, Gunicorn+Uvicorn workers, .dockerignore hygiene, and image-size checks (~600-700 MB target). - Managed state migration: Move local vector stores, Redis history, and tracing to managed cloud services with only connection-call changes. - Vector store tuning: Benchmark precision@k against exact search, sweep hnsw_ef, and apply scalar quantization with rescoring. - Use Case: You have a working RAG chatbot running locally with Qdrant and want to ship it on AWS. The Skill walks you through API extraction, ECR push, EC2 deployment, smoke testing, and setting scaling-ladder triggers. ## Quick Start Ask the assistant to take your local LLM prototype and deploy it as a production containerized service on your cloud account.
