Document segmentation / chunking

Break documents into retrievable pieces.

Sending an entire large document to every query can:

• waste tokens
• reduce retrieval precision
• increase latency
• dilute relevant information

#### Fixed-size chunking

Split by approximate size.

Simple but may cut semantic units.

#### Semantic chunking

Split according to meaning.

#### Hierarchical chunking

Document
 ├── Chapter
 │    ├── Section
 │    └── Section
 └── Chapter

Useful for structured documents.

• Amazon Bedrock Knowledge Bases chunking
• AWS Lambda
• custom processing pipelines

There is no universally best chunk size.

Chunking depends on:

• document structure
• query patterns
• model context
• retrieval quality

---