A developer has created a local Retrieval-Augmented Generation (RAG) system designed to make research lab notes and slides searchable. The system utilizes multiple AI models, including Gemma 4.31B for slide descriptions and Qwen3 Embedding 4B for text embedding, running on AMD hardware. A larger model, DeepSeek-v4.1-Flash-1017, processes queries and generates answers using retrieved excerpts, with the entire setup hosted on Nvidia DGX Spark systems. The system aims to provide researchers with precise answers traceable to specific files and line or slide numbers, with indexing and updates scheduled daily. AI
IMPACT Enables researchers to query and retrieve information from unstructured lab data, potentially speeding up knowledge discovery.
RANK_REASON Developer-built RAG system using existing models and hardware.
Read on Mastodon — mastodon.social →
- AMD
- DeepSeek-v4.1-Flash-1017
- DGX Spark
- Gemma 4.31B
- Japan
- Nvidia
- Ollama
- Open-WebUI
- qdrant
- Qwen3 Embedding 4B
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →