Marvin-LiJunZan Paper RAG — Literature Knowledge Base & Multi-Paper QA
Paper RAG transforms your collection of research papers (PDFs, Markdown notes, BibTeX libraries) into a queryable, evidence-grounded research knowledge base.
Capabilities
Cross-Paper Synthesized QA:
- Ask questions across dozens of papers simultaneously (e.g., "How do different studies handle the water-to-binder ratio in alkali-activated slag?").
- Returns responses strictly grounded in retrieved chunks with exact paper citation keys and section references.
Long-Term Topic Knowledge Accumulation:
- Maintains an incremental index of papers in
corpus/. - Supports adding new papers, tagging by sub-topics, and tracking conflicting findings across publications.
- Maintains an incremental index of papers in
Evidence Matrix Aggregation:
- Automatically gathers comparative data (metrics, datasets, parameter ranges) across papers into structured markdown tables.
Quick Usage
Run the bundled lightweight Python RAG engine:
# Index a directory of papers or notes
python scripts/paper_rag.py index --dir "path/to/papers" --db "paper_index.json"
# Query the literature base
python scripts/paper_rag.py query --db "paper_index.json" --query "How is compressive strength affected by curing temperature?"
Query Modes
- Synthesize Mode: Merges viewpoints from multiple authors, highlighting agreements and controversies.
- Compare Mode: Constructs a head-to-head table of methodologies, assumptions, and outcomes.
- Fact-Check Mode: Locates whether any paper in your corpus supports or contradicts a specific claim.