A production RAG assistant, end to end
From a folder of PDFs to an assistant your team actually trusts.
The idea
The RAG lesson shows the core in 22 lines. A real assistant needs much more: document ingestion and chunking, a vector index, an API, a simple UI, citations, access control, and an eval set that runs on every change.
This project builds all of it as a multi-part series, with each part ending in working code.
What the lesson will build
- Ingestion: parse, chunk and embed a document folder
- Retrieval API with citations and access rules
- A minimal web UI
- An eval suite and a deploy
Key ideas
- Chunking strategy
- Hybrid search and reranking
- Evals as a release gate
- Serving and monitoring
The video
- Long form: A series: each episode adds one component to an architecture diagram that fills in over the parts.
- Short: Part by part, a real assistant gets built.
When it’s published, the code will live in build-projects/ and this page will link to it.