Mini course
Build a RAG Assistant
One project, start to finish.
Build a company-policy Q&A assistant with retrieval-augmented generation, from chunking to evaluation.
13 topics2.3 h hands-on4 h in depthLifetime access
What you'll be able to do
- Chunk, embed and index documents
- Answer questions with cited sources
What's inside
Module 4 · Project 2: Company Policy Q&A Assistant (RAG — Retrieval-Augmented Generation)13 topics · 2.3 h
4.1 Understand the problem1 topics · 0.2 h
- Why RAG and how we will judge it10 min
4.2 Data and test set1 topics · 0.2 h
- Collect documents and write the test questions10 min
4.3 Prepare the documents2 topics · 0.3 h
- Read PDFs and split into chunks10 min
- Embeddings and the vector database10 min
4.4 Generate the answer2 topics · 0.3 h
- Answer prompt with citations and 'I don't know'10 min
- Chat memory and context limits10 min
4.5 Use a framework1 topics · 0.2 h
- The same RAG pipeline in LangChain and LlamaIndex10 min
4.6 Evaluate1 topics · 0.2 h
- Retrieval score, correctness, faithfulness and refusals10 min
4.7 Improve2 topics · 0.5 h
- Query rewriting, chunk size, re-ranking and metadata filters20 min
- Glimpse: advanced RAG patterns10 min
4.8 Safety2 topics · 0.3 h
- Input guardrails: personal data and harmful requests10 min
- Prompt injection and data access10 min
4.9 Deploy1 topics · 0.2 h
- Gradio chat with sources and logging10 min