athrael-soju/Snappy

🐊 Snappy's unique approach unifies vision-language late interaction with structured OCR for region-level knowledge retrieval. Like the project? Drop a star! ⭐

51
/ 100
Established

Implements region-level document retrieval by mapping ColPali's patch-level similarity scores to OCR-extracted bounding boxes via spatial coordinate propagation, enabling fine-grained retrieval without additional training. Operates as a two-stage pipeline: efficient candidate prefetching using mean-pooled embeddings followed by full-resolution region reranking with interpretability heatmaps. Built on FastAPI and Next.js with Qdrant vector storage, DeepSeek-OCR-2 integration, and Docker deployment for production RAG systems.

No Package No Dependents
Maintenance 10 / 25
Adoption 9 / 25
Maturity 15 / 25
Community 17 / 25

How are scores calculated?

Stars

83

Forks

15

Language

Python

License

MIT

Last pushed

Feb 09, 2026

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/vector-db/athrael-soju/Snappy"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.