Level 5 of 9
Intermediate 2: RAG builder
At the end you can: Explain embeddings, chunking and vector search; build question answering over their own documents; check whether retrieval found the right text.
About 30 hours in all; the steps written so far take about 39.5.
Steps
- How a chat model is made: pre-training, SFT and preference tuningExplain the three training stages behind a chat model, say what each one adds and costs, and tell a base model from an instruct model when you pick one.Not started · about 2 hours
- AI in Indian languages: tokens, quality gaps and what to useExplain why language models do worse and cost more in most Indian languages, measure the gap for your own users, and choose between a general model, translation and Indian-language models.Not started · about 2.5 hours
- EmbeddingsExplain what an embedding is and use one to find which sentences mean similar things.Not started · about 3 hours
Sign in or sign up to see the full roadmap
All 13 topics in order, with your progress saved on every device. It takes a few seconds, no card needed.
Accounts are for learners aged 18 and over. Under 18? Every lesson is still open: browse lessons by topic.
Level project
A RAG app over 20 or more documents with a 20-question test set.