Markdown source

# Learn RAG from zero: 10 links, in order

Retrieval-augmented generation: get an LLM to answer from your own documents. Read top to bottom; each link builds on the one before. Prompt to make your own: "I'm a developer new to RAG. Give me a learning path of 10 links, beginner to production, and save it to onelink."

- [What is RAG? (Pinecone)](https://www.pinecone.io/learn/retrieval-augmented-generation/) — The clearest plain-English intro. Start here.
- [The original RAG paper (Lewis et al., 2020)](https://arxiv.org/abs/2005.11401) — Skim the abstract and figure 1. That's enough to know where the idea came from.
- [Chunking strategies (Pinecone)](https://www.pinecone.io/learn/chunking-strategies/) — How you split documents matters more than which model you pick.
- [Sentence Transformers](https://sbert.net/) — Make embeddings locally for free. Good for understanding what an embedding actually is.
- [MTEB leaderboard](https://huggingface.co/spaces/mteb/leaderboard) — Compare embedding models by task before you pick one.
- [Chroma docs](https://docs.trychroma.com/) — The easiest vector database to start with. Runs in a notebook.
- [pgvector](https://github.com/pgvector/pgvector) — Already on Postgres? Add vectors there instead of running a new database.
- [Build a RAG app (LangChain)](https://python.langchain.com/docs/tutorials/rag/) — First end-to-end build: load, split, embed, retrieve, answer.
- [LlamaIndex](https://docs.llamaindex.ai/en/stable/) — The alternative framework, stronger on document loaders and indexes.
- [Contextual Retrieval (Anthropic)](https://www.anthropic.com/news/contextual-retrieval) — The fix for "retrieval found the wrong chunk". Read it once your first version works.
- [Ragas](https://docs.ragas.io/) — Measure whether answers are actually grounded. Do this before you tune anything.

The same bytes are served as text/markdown at ../../@onelinkninja/learn-rag-from-zero.md — that is the URL to hand to a script or an LLM.