Semantic Chunking

~15 min read

Group sentences into a chunk as long as consecutive segments stay semantically similar (via cosine similarity of their embeddings), starting a new chunk exactly where the similarity drops significantly.

Semantic Chunking is a Pro topic

Sign in, then upgrade to Pro or Power to unlock this topic and the full AI Engineering curriculum.

Key points

  • Chunks based on meaning: consecutive segments stay in one chunk while their embeddings remain highly similar
  • A significant drop in cosine similarity between consecutive segments signals a topic shift and starts a new chunk
  • Preserves the natural flow of language and complete ideas — unlike fixed-size chunking
  • Richer, more topically-coherent chunks generally improve retrieval accuracy and downstream response quality
  • The similarity-drop threshold isn't universal — it varies by document type and typically needs per-corpus tuning