Agent Deployment
Every deployment path the AISL workshops have actually taken an agent through — FastAPI on Railway, AWS Lambda, Cloudflare Workers, Vercel, and vLLM on rented GPUs.
A notebook nobody can reach is a demo. The deployment workshops take the same
FAQ agent — one search tool over the DataTalks.Club Zoomcamp FAQ — through
a different platform each time, so the platforms can be compared on real
experience rather than marketing.
The reference path: notebook to production
End-to-End Agent Deployment is the starting point: FastAPI backend, vanilla-JS frontend with SSE streaming, Docker, Railway, GitHub Actions CI/CD. Every step runs alongside a coding agent, with the prompts included verbatim.
Owning the runtime on AWS
Agent Deployment on AWS Lambda with a Custom Runtime strips FastAPI out and replaces it with a custom Lambda runtime: routing by method and path, SSE through a thread, one-command deploys, and a separate AWS account so the agent cannot touch production resources.
The edge platforms, tested
Two workshops ask the same question — can a JavaScript-first edge platform carry a Python-born agent?
- A FAQ Agent on Cloudflare Workers: Worker for UI and API, Workers AI for embeddings and generation, Vectorize for vectors, with pricing math and cleanup included.
- A FAQ Agent on Vercel: the Vercel AI SDK and AI Gateway, Upstash semantic search, a Python backend attempt, and an explicit comparison of every version built in the session.
Small and free
Vector Search with SQLite and Turso changes the data layer instead of the platform: local ONNX embeddings, a vector index in one SQLite file, hosted on Turso — no Postgres, no rented disk, no server billed while idle.
Self-hosting open models
Serving Open Models with vLLM on RunPod covers the other side of deployment: renting a GPU by the hour, serving a quantized DeepSeek model with vLLM behind an OpenAI-compatible API, running a tool-calling agent against it, and shutting the bills off when you stop.
Where to go next
- The agent you are deploying: Agents
- The search layer most of these workshops swap in: Vector Search
- Keeping the deployed agent safe and observable: Guardrails and Evaluation and Monitoring
Keep building with a membership
Every topic guide is free to read. Members get structured courses, live workshops, and feedback on real projects.