Local RAG system lets lab researchers query notes and slides with traceable citations

A research team built a local retrieval-augmented generation service that answers questions about their lab's Markdown notes and PowerPoint slides. Gemma 4 31B generates text descriptions of rendered slides so that figure content becomes searchable, while Qwen3-Embedding-4B embeds both source text and those descriptions into a Qdrant index. Answers are produced by DeepSeek-V4.1-Flash-1017 from up to four retrieved excerpts, and the service lists every supplied excerpt with its file path and line or slide number. In a November 2026 test over 143 files, both sample questions in Japanese and English returned the expected path with a supporting citation. The author notes that valid citation IDs only locate passages, so a person must still verify that each passage supports the claim it is cited for.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in