GyaanStack. Blog

Articles / ragvideosfree-resources

RAG end to end in free videos

Ten free videos that take you from embeddings and vector search to hybrid search, self-correcting RAG and evaluating the whole pipeline.

By Raghunath · · 5 min read

Most RAG tutorials on YouTube build the same demo: split a PDF, embed it, ask a question. That gets you a working prototype and very little about why it gives wrong answers on real documents.

This list follows the RAG, end to end path on GyaanStack Videos, with one video from the Evaluating LLM apps path added at the end. It goes from the maths of embeddings to production problems: hybrid search, re-ranking, pipelines that fix their own bad retrievals, and how to measure all of it. It's ten videos and about seven and a half hours.

Part 1: how retrieval works#

1. Retrieval Augmented Generation (RAG) Explained: Embedding, Sentence BERT, Vector Database (HNSW)Umar Jamil · 49 minEmbeddings and vector search from the maths up, including HNSW, the graph index most vector databases use. Watch this first if "cosine similarity" is still a phrase you nod along to.

2. RAG Explained: From BM25 to Agentic RAGStanford Online · 1 h 16 minLecture 3 of Stanford's CS329Z. It covers the whole range in one lecture, from keyword search with BM25 to agentic RAG, where the model decides when and what to retrieve. This gives you the map for the rest of the list.

3. How modern search engines work – Vector databases explained!AI Coffee Break with Letitia · 14 minHow vector databases make search fast. Short, and a good break after two long videos.

Part 2: build a pipeline that holds up#

4. A guide to building Retrieval Augmented Generation (RAG) pipelines that actually workNeural Breakdown with AVB · 17 minThe design choices in a working pipeline, step by step. After this one, build a small version yourself before going on. My PDF question-answering bot is a starting point that runs without an API key.

5. Improving RAG Systems with Hybrid Search and Re-ranking with LanceDBJason Liu · 15 minHybrid search combines keyword and vector results, and a re-ranker reorders the top hits. Of everything on this list, these two are the cheapest changes to try on a pipeline that returns the wrong chunks.

Part 3: pipelines that check themselves#

Both videos in this part are from CampusX and run over an hour each. Watch them when your basic pipeline works and you want to handle the cases where retrieval comes back with nothing useful.

6. Advanced RAG: How Corrective RAG (CRAG) Solves Traditional RAG ProblemsCampusX · 1 h 15 minCorrective RAG grades the retrieved documents and takes another route, such as a web search, when they aren't good enough.

7. Self-RAG Tutorial: How to Make Your AI Fact-Check ItselfCampusX · 1 h 8 minSelf-RAG has the model check its own answer against the sources. It costs more calls per question, so it suits cases where a wrong answer is expensive.

Part 4: measure it#

Without measurement, every change in parts 2 and 3 is a guess. These three videos cover how to tell whether a change helped.

8. Introduction to LLM Evaluations – Model Evals vs Application EvalsCampusX · 24 minFrom the Evaluating LLM apps path. It separates benchmark scores for a model from tests of your own application, which is the kind you need for RAG.

9. Modern Information Retrieval Evaluation In The RAG EraHamel Husain · 57 minHow to measure retrieval quality on its own, before the LLM writes anything. If retrieval is wrong, no prompt will fix the answer.

10. Evaluating RAG: Testing the Generator & Full Pipeline with the RAG TriadCampusX · 1 h 21 minTesting the generation step and the whole pipeline with the RAG triad: whether the retrieved context is relevant, whether the answer is grounded in it, and whether the answer addresses the question.

If you only have one evening#

Watch video 2 (1 h 16 min) for the map, video 5 (15 min) for the fixes that help most often, and video 9 (57 min) for measuring retrieval. That's about two and a half hours and covers the parts of RAG most tutorials skip.

Who this is for#

This list assumes you've called an LLM API and know roughly what an embedding is. If you haven't, watch the first two videos of the how LLMs work weekend plan before starting here. If you've already shipped a RAG system, start at part 4 and measure what you have before adding anything new.

A plan for two weeks of evenings#

Days Watch Then do
1–3 Videos 1–3 Embed 20 short texts and run a few searches by hand
4–5 Video 4 Build a small pipeline over one document you know well
6 Video 5 Add BM25 next to vector search and compare the top 5 results
7–9 Videos 8–9 Write 15 questions with the expected source passage, and measure recall@5
10–12 Video 10 Add answer checks: grounded in the context, answers the question
13–14 Videos 6–7 Add a corrective step and see whether your numbers move

I put the evaluation videos before the self-correcting ones in this plan on purpose. You can only tell whether CRAG or Self-RAG helped if you already have the numbers from before.

For a one-page reference while you build, keep the RAG cheat sheet open. The written courses and papers for each stage are in the free RAG resources post and the GyaanStack RAG subject. The full video path is at videos.gyaanstack.com.