PaperTrail: The Research Memory Agent

Researchers and students read dozens of papers but struggle to connect the dots weeks later. They often remember what a method was, but forget which specific paper proposed it. Highlights and notes are scattered across PDFs and notebooks, and standard keyword search fails because it finds words, not concepts or relationships between papers.

A research brain that ingests PDFs and notes into a unified workspace. Instead of treating text as blobs, it automatically builds a visual knowledge graph connecting papers, linking entities like Authors, Methods, Datasets, and Metrics. Users can ask complex questions like 'Which papers compare GraphRAG with embedding-only retrieval, and what tradeoffs do they report?' and receive a structured, cited answer.

• Upload PDFs, notes, links, or pasted text
• GPT-4o extracts entities and relationships from the content
• The knowledge graph updates as new papers are added
• Ask questions — GraphRAG retrieves nodes and passages, then returns cited answers

• Entity extraction — GPT-4o reads raw text and extracts structured nodes (Methods, Datasets) to populate the graph
• GraphRAG — hybrid retrieval combining vector search with graph traversal to answer reasoning questions that simple chatbots miss

• Stack: Python, React, Cytoscape.js
• AI & Data: GPT-4o, Neo4j, Qdrant
• Dev: Cursor, GitHub Copilot