Roadmap
Retrieval-augmented generation (RAG)
From your first chat-with-your-PDF demo to a retrieval system you can measure and trust.
Who it's for
Developers building assistants, search or support bots on top of their own documents.
Where it leads
- AI engineer
- Search engineer
- LLM engineer
By the end you can
- Choose an embedding model and vector store with reasons
- Build an end-to-end RAG pipeline
- Improve retrieval with chunking, hybrid search and reranking
- Measure retrieval and answer quality separately
- 1
The idea, from the source
Why retrieval beats stuffing a model with facts, and what the original architecture did.
- 2
Embeddings and vector search
Pick an embedding model, store vectors, and understand similarity and indexes.
- 3
Build your first pipeline
Load, chunk, embed, retrieve and generate, end to end, in an afternoon.
Do this: Build a bot that answers questions about one real document set you own, with citations.
- 4
Make retrieval good
Chunking, hybrid search, reranking and query rewriting are where quality is won.
- 5
Measure it
Separate retrieval quality from answer quality, and test every change against a fixed set.
Do this: Measure recall@5 on 30 real questions, then improve it with one change at a time.
Learning this with others is easier.
Ask questions when you're stuck, find people on the same path, and track your progress on DevLearn.