Vertex

Chunking strategies

Chunk size decides what a match can even mean. Too small and passages lose context; too large and the relevant sentence gets drowned.

Lesson 2.114m 2sAdvanced7.7k students

Chapters

14:02
  1. 0:00Introduction to Chunking Strategies in RAG
  2. 1:52Detailed Tutorial on Various Chunking Methods
  3. 2:25Setup Instructions for Chunking Environment
  4. 3:01Code Walkthrough for Character Text Splitting
  5. 6:01Implementing Recursive Character Text Splitting
  6. 7:00Exploring Document Text Splitting Techniques
  7. 8:26Introduction to Semantic Chunking with Embeddings
  8. 9:45Advanced Agentic Chunking for Optimised Grouping
  9. 13:50Conclusion

What this lesson covers

  • Split on structure rather than a fixed character count
  • Use overlap to avoid cutting an idea in half
  • Keep each chunk understandable on its own

This lesson sits in Ingestion and Indexing, part of Retrieval-Augmented Generation from Scratch. It assumes what came before it and leads directly into the next lesson in the module.

In this lesson you will

  • Split on structure rather than a fixed character count
  • Use overlap to avoid cutting an idea in half
  • Keep each chunk understandable on its own

Resources

Previous lessonSimilarity metrics and thresholds1.3 · 7m 53sNext lessonMetadata and filtered retrieval2.2 · 12m 24s