From localhost to Real World: Engineering Production-Ready Agentic Systems
summary
Concrete claim: a production agent pipeline that scales to 300,000 users per hour with retry, timeout and guardrail patterns implemented. The demo repo wires a Streamlit UI to FastAPI and Streamlit backed by a Redis/Celery queue and model calls (Gemini in the demo) with schema validation via Pydantic, infra-as-code using Terraform, and logging into LangFuse observability. The system runs asynchronously, enforces guardrails and CI/CD deployment, and the demo executes end-to-end.
AI-written summary · about
chapters (9)
- 0:00Building production AI agents
- 1:41Local dev fails in prod
- 4:29Four pillars of AI agents
- 7:16Guardrails and data validation
- 8:42Async workflows for scalability
- 9:24Security and scalability in AI
- 10:50Infrastructure beyond LLM
- 13:45Demo: production multi-agent system
- 17:56Closing remarks & Q&A
