Description
GreyMatter is a RAG-powered scientific discovery platform built for the Claude Solv-A-Thon hackathon. It transforms decades of dark matter research literature into living debates: retrieving, comparing, and transparently arguing across experiments with citations and uncertainty quantification. Using hybrid retrieval (dense embeddings + BM25), Claude-powered debate synthesis, and interactive consensus graphs, it makes cutting-edge astrophysics research accessible to both experts and the public.
Tech Stack
Next.js 15TypeScriptTailwind CSSAnthropic Claude AIGoogle Gemini AIPineconeLangChainPrisma ORMPostgreSQLBetter AuthVercel AI SDKGroqReact Graph Visshadcn/uiFramer Motion
Highlights
•Hybrid RAG retrieval combining dense embeddings with BM25 keyword search, optional cross-encoder reranking, and MMR diversification to avoid duplicate results
•Claude-powered debate synthesis generating expert summaries with methods and caveats, layperson analogies, embedded glossaries, and argument maps (supports/refutes/uncertain)
•Interactive consensus and contradiction graph visualization using React Graph Vis, mapping support, conflict, and uncertainty relationships across research papers
•Uncertainty quantification scoring each claim on source weight (peer-reviewed > preprint), recency, agreement count, method quality, and statistical strength
•Voice-powered research interface with text-to-speech synthesis, enabling hands-free exploration of dark matter literature and debate maps