RAG Learning Path 3 YouTube Resources Beginner → Advanced

Learn RAG from These Resources

A focused learning path for Retrieval-Augmented Generation: understand the complete pipeline, build it yourself, then learn how to improve retrieval quality, reduce hallucinations, evaluate results, and explain the system confidently in interviews.

RAG in one minute

Retrieval-Augmented Generation (RAG) gives an LLM access to external knowledge at query time. Instead of expecting the model to know everything, your application retrieves relevant passages from your documents or knowledge base and sends that context with the user's question to the model.

1. Ingest

Load PDFs, webpages, docs, databases or other knowledge sources.

2. Index

Split content, create embeddings and store searchable representations.

3. Retrieve

Find the chunks most relevant to the user's query.

4. Generate

Ask the LLM to answer using the retrieved context and cite sources when possible.

Start Here

3 YouTube resources to learn RAG

Use them in sequence. Do not only watch—code along, take architecture notes, and rebuild the pipeline without copying.

1Crash Course

Complete RAG Crash Course With LangChain in 2 Hours

A strong first pass through the RAG workflow. Watch it to connect document loading, splitting, embeddings, vector storage, retrieval and LLM generation into one mental model.

Watch on YouTube ↗

Focus on

End-to-end RAG flow and how LangChain components connect.

While watching

Draw the architecture and write down where data changes form: text → chunks → embeddings → retrieved context.

After watching

Build a tiny PDF Q&A app with citations and inspect the chunks that were actually retrieved.

2Playlist

CampusX RAG Playlist

Use this as your deeper structured learning track. A playlist format is useful because you can revisit individual building blocks rather than treating RAG as one monolithic topic.

Open Playlist ↗

Focus on

Chunking, embeddings, vector stores, retrievers and how design choices affect answer quality.

While watching

Create a glossary for similarity search, top-k, metadata filtering, reranking, grounding and hallucination.

After watching

Compare at least two chunk sizes and retrieval settings on the same question set.

3From Scratch

Learn RAG From Scratch — Python AI Tutorial from a LangChain Engineer

Use this to reinforce the fundamentals from a build-first perspective. The goal is to understand what every stage does rather than memorizing framework syntax.

Watch on YouTube ↗

Focus on

The underlying Python flow and the responsibilities of each RAG component.

While watching

Pause before each step and predict what input/output the next component should receive.

After watching

Rebuild the same concept using your own documents, prompts and evaluation questions.

Learning Roadmap

What you should learn, in order

1. RAG fundamentals

Why RAG exists, private/current knowledge, grounding, indexing vs retrieval, and RAG vs fine-tuning.

2. Document ingestion

PDF/web loaders, cleaning, metadata, document structure and preserving source information.

3. Chunking

Chunk size, overlap, semantic boundaries, recursive splitting and why poor chunking damages retrieval.

4. Embeddings

How text becomes vectors, semantic similarity, model choice and matching query/document embeddings.

5. Vector stores & retrieval

Similarity search, top-k, metadata filters, vector databases and retriever abstractions.

6. Prompting & generation

Context injection, answer-only-from-context instructions, citations and handling insufficient evidence.

7. Advanced retrieval

Hybrid dense + keyword retrieval, reranking, query rewriting, multi-query retrieval and parent-document strategies.

8. Evaluation & production

Retrieval relevance, answer faithfulness, test datasets, latency, cost, caching, observability and failure analysis.

Official References

Read these after the videos

LangChain Retrieval Docs ↗

Understand retrieval building blocks and the differences between 2-step, agentic and hybrid RAG architectures.

LlamaIndex Introduction to RAG ↗

A clear explanation of how data is indexed, filtered to relevant context and then supplied to an LLM.

RAG Evaluation Tutorial ↗

Move beyond “it looks good” by creating datasets and measuring answer and retrieval quality.

Original RAG Paper ↗

The foundational paper: Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks.

Build While You Learn

Your first portfolio RAG project

Build a domain-specific document assistant instead of another generic “chat with PDF” demo. Examples: company-policy assistant, technical-documentation assistant, research-paper assistant, product-support knowledge bot or job-description/resume knowledge assistant.

MVP

Upload documents → chunk → embed → store → retrieve → generate an answer with source references.

Upgrade

Add metadata filters, hybrid search, reranking, conversation history, query rewriting and “I don't know” behavior.

Prove quality

Create 20–50 test questions and compare retrieval hit rate, answer correctness, faithfulness, latency and cost before/after changes.

RAG interview checklist

You should be able to explain these without depending on framework-specific code:

✓ RAG vs fine-tuning and when to use each

✓ Why chunk size and overlap matter

✓ Embeddings and semantic similarity

✓ Vector search vs keyword/BM25 search

✓ What top-k means and how to tune it

✓ Why reranking improves retrieval

✓ How to reduce hallucinations

✓ How to evaluate retrieval separately from generation

✓ What hybrid search is

✓ How metadata filtering helps

✓ How citations/source attribution work

✓ Latency, token cost, caching and observability

A simple 7-day RAG learning plan

Day 1: Learn the RAG architecture and complete the first crash course.

Day 2: Study loaders, chunking and embeddings. Experiment with chunk sizes.

Day 3: Learn vector stores, similarity search, top-k and metadata filtering.

Day 4: Build a complete RAG application with your own documents.

Day 5: Add hybrid search/reranking and compare retrieval quality.

Day 6: Create an evaluation set; test correctness, faithfulness, latency and failures.

Day 7: Clean the GitHub README, add architecture, screenshots, metrics and prepare interview explanations.

Start with Video 1 ↗ Explore More Resources