AI Engineering

Retrieval-Augmented Generation from Scratch

Build a RAG system that actually retrieves the right thing: embeddings, chunking, hybrid search, reranking, and grounded answers.

  • Advanced
  • 1h 57m
  • 4 modules
  • 12 lessons
  • 9.1k students
Start course

Taught by Priya Raman · LLMs, RAG, Prompt engineering, Evaluation

Cover image for Retrieval-Augmented Generation from Scratch

What you’ll learn

  • Understand embeddings

    Know what a vector encodes, what it does not, and which similarity measure to use.

  • Chunk content well

    Split documents so retrieved passages are self-contained and carry useful metadata.

  • Improve retrieval quality

    Combine keyword and vector search, then rerank, and measure whether it helped.

  • Ground every answer

    Cite sources, and make the system say it does not know instead of inventing one.

Course content

4 modules · 1h 57m

  1. 01Embeddings and Vector SearchThe retrieval primitive: turning text into vectors and finding the near ones quickly.21m
    1. 1.1What an embedding actually encodesFree9m
    2. 1.2Vector databases and indexes4m
    3. 1.3Similarity metrics and thresholds8m
  2. 02Ingestion and IndexingGetting your content into the index in a shape that retrieves well and stays current.32m
    1. 2.1Chunking strategies14m
    2. 2.2Metadata and filtered retrieval12m
    3. 2.3Keeping the index fresh6m
  3. 03Retrieval QualityWhere most RAG systems are won or lost: combining retrieval methods, reranking, and measuring the result.24m
    1. 3.1Hybrid keyword and vector search7m
    2. 3.2Reranking retrieved candidates5m
    3. 3.3Evaluating retrieval12m
  4. 04Grounded GenerationTurning retrieved passages into an answer that is traceable, honest, and measurable.40m
    1. 4.1Grounding answers with citations7m
    2. 4.2Handling the no-answer case22m
    3. 4.3End-to-end RAG evaluation11m
Start with the first lesson