RAGpack: Document-to-Vector-Store Ingestion Library

Personal: RAG & Retrieval Systems · 2026

Turns a folder of documents into a searchable Qdrant index in one line: GPU-optional embeddings, optional GPU OCR for scanned PDFs, and fail-closed device selection. Installable from source.

installable

from source; CPU / GPU / OCR extras measured

Py 3.10–3.12

CI matrix, fully offline test suite measured

fail-closed

device="cuda" never silently drops to CPU measured

Python

Qdrant

Embeddings

GPU OCR

Context

Every RAG project starts by re-solving the same unglamorous problem: get a pile of documents into a vector store cleanly, without silent data loss. RAGpack is that step, extracted into a library so it stops being rewritten.

What I built

One-line ingestion: point it at a directory, get a searchable Qdrant index, with embeddings on CPU or GPU.
Optional GPU OCR for scanned and image-based PDFs.
Fail-closed device selection: ask for cuda and it will not quietly fall back to CPU and pretend everything is fine; it tells you.

Where it’s used

It is the ingestion engine under the RFC-corpus RAG and the blockchain vertical on this site: the same code, installable from source, with an offline test suite that runs on Python 3.10 through 3.12 in CI.