evidence-preserving-chunking

Builds and benchmarks document segmentation for retrieval-augmented systems with exact source provenance: character and model-token windows, recursive, structural, semantic, clustered, hierarchical and hybrid splitters, contextual retrieval, late chunking, frozen-dataset evaluation with ranking metrics, bounded caching, streaming, and HTML/SVG reporting, all as compact inline Python modules. Use when the user says "chunk documents for RAG", "split PDFs or Markdown for embeddings", "compare chunking strategies", "tune chunk size or overlap", or "evaluate retrieval quality". Not for prompt engineering, model fine-tuning, or operating a live production index.

Version

v1.0.1

Size

42 KB

Updated

2026-10-06

Your client verifies the signature before reading. This explicit manual read is available on every plan. Free automatically loads five chosen skills; paid plans automatically load relevant skills across the library.