Snappy
What it is
Snappy is a vision-first document retrieval system that combines vision-language models (ColPali) with structured OCR. It uses a unique 'patch-to-region relevance propagation' technique to map visual embeddings to specific OCR-detected regions, allowing for precise retrieval of paragraphs, tables, and figures rather than entire pages.
The problem it solves
Traditional document retrieval systems either return entire pages (too much noise) or lack the semantic understanding of visual layouts (OCR-only).
What you could use it for
- Retrieving specific tables or figures from large PDF documents
- Visual RAG (Retrieval-Augmented Generation) with high precision
- Generating interpretability heatmaps for document query matching
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.