Overview
The National Level Claude Solv-A-Thon, hosted at IIIT Nagpur, challenged developers across the nation to build cutting-edge applications using Anthropic’s Claude LLMs. Out of numerous competitive submissions, our project GreyMatter won the 2nd Place position.
GreyMatter is a RAG-powered scientific discovery platform designed for astrophysics literature. It transforms decades of dark matter research literature into living debates—retrieving, comparing, and transparently arguing across experimental findings with full citations and uncertainty quantification.
The Problem
Astrophysics and dark matter literature are filled with contradictory observations and hypotheses (e.g., DAMA/LIBRA annual modulation claims vs. XENONnT exclusion limits). Researchers and the public struggle to synthesize these complex disputes. Existing search engines only retrieve papers, without highlighting consensus or pointing out contradictions.
Our Solution: GreyMatter
GreyMatter resolves this by:
- Hybrid Retrieval: Blending dense embeddings with BM25 keyword search over a specialized dataset of astrophysics preprints.
- Debate Synthesis: Using Claude to construct structured arguments between different research papers (Supports, Refutes, or remains Neutral).
- Consensus Graphs: Building an interactive network visualization mapping relations between research papers and specific astrophysics claims.
- Uncertainty Scoring: Rating claims based on recency, method quality, and statistical strength.
Key Takeaways & Impact
Winning 2nd place at IIIT Nagpur validated our approach of using LLMs not just as copywriters, but as reasoning engines that can synthesize complex scientific disputes. The event highlighted the power of Anthropic’s Claude model in handling long-context scientific papers with high reasoning precision.