Mindket
English
English简体中文
Sign inSign up
StoreMCPSkillsCalculatorLearnPriceBlog
Discover
Research308Writing59Business95Legal13Finance & accounting33Marketing & SEO172Developer1,379Productivity382Education81Companions14
Help and support
Language
English简体中文
AdvancedFree

RAG course: build retrieval-augmented generation

Mindket editors · Updated · 4 lessons · 41 min

A black banner titled "RAG" over the line "Embeddings, chunking, retrieval and evals".

This free RAG course for developers has four lessons. You build a retrieval-augmented generation pipeline with citations, learn how embeddings and pgvector work, tune chunking with contextual retrieval, hybrid search and reranking, and measure the result with Ragas. Every step follows vendor and project documentation, and you need Python and basic API experience.

What you'll be able to do

  • Build a RAG pipeline that loads, splits, embeds, retrieves and answers with citations.
  • Choose an embedding model and a dimension, then store and query vectors in Postgres with pgvector.
  • Pick a chunk size, split by structure, add context to chunks and combine keyword and vector search.
  • Score retrieval and answers with context recall, context precision, faithfulness and answer relevancy.
All courses

Who it's for

For developers who write Python and have called an API before. You need Python 3 and API keys for an embedding model, Claude and a judge model, plus a Postgres server with pgvector for lessons 2 and 3.

Lessons

  1. 1RAG tutorial: build a retrieval pipeline step by stepA RAG tutorial for developers: split documents, embed them with Voyage AI, retrieve by similarity and have Claude answer with citations.9 min
  2. 2What are embeddings? Models, dimensions and pgvector for developersWhat are embeddings and how do you use them? Pick a model and a dimension, compare vectors with cosine similarity and store them in Postgres.10 min
  3. 3RAG chunking strategies: chunk size, overlap, contextual retrieval and hybrid searchRAG chunking strategies for developers: choose a chunk size and overlap, split by structure, add context to each chunk and combine BM25 with vectors.11 min
  4. 4How to evaluate RAG: retrieval metrics, faithfulness and a small test setHow to evaluate RAG: build a small test set, measure retrieval with context recall and precision, and score answers for faithfulness with Ragas.11 min

Frequently asked questions

Is this RAG course free?

Yes. Every lesson is free to read. To follow along you need API keys for an embedding model, Claude and a judge model, and those providers may bill for their use.

Do I need LangChain or LlamaIndex?

No. Lesson 1 retrieves with a few lines of NumPy and calls the model directly. It borrows LangChain's text splitter for chunking, and lessons 3 and 4 use LangChain's splitters and Ragas.

Which vector database does the course use?

Postgres with the pgvector extension. Lesson 2 covers its distance operators, its HNSW and IVFFlat indexes, its dimension limits and filtered queries, and lesson 1 starts with vectors held in memory.

Which versions of the tools do the lessons follow?

The documentation as it stood on 5 November 2026, including pgvector 0.8.7, the Voyage 4 model series and Ragas 0.4.3, the newest release on PyPI. The Ragas pages it relies on are dated 9 December 2025, so pin the version you install.

What should I learn after this course?

The advanced prompt engineering course is next, the first course of the Master prompting path. Its first lesson teaches a model with examples, and the path goes on to reasoning models and prompt optimization.

Share this course

See what a retrieval pipeline can power

Browse the agents on Mindket, each listed with a preview, to see what an agent can do with your own documents.

Browse agents

All courses · All articles