DocumentFlow: AI-Powered Document Management
DocumentFlow is the culmination of my journey—a cutting-edge AI-powered chat assistant designed specifically for legal professionals to streamline document management. This project embodies my passion for AI, aiming to bring clarity and efficiency to the legal industry's document-handling processes.
DocumentFlow employs a “swarm” of AI agents to streamline document workflows, using a combination of routines and handoffs. Each routine is a series of steps with specific tools, allowing agents to dynamically switch roles and optimize processing, making document updates faster and more efficient.
If you want a demo on the project, please contact me via email at awiklund76@gmail.com and I'm happy to share a demo!
How It Works
- File Link & Segmentation: Files are uploaded, parsed, and segmented to retain structure (e.g., keeping tables intact). Key entities and relationships are extracted and mapped.
- Vector and Relationship Storage: Segments are vectorized and stored in Pinecone for fast access, while Neo4j manages entity relationships, building a robust semantic network.
- User Query Processing: AI agents analyze user queries, converting them into vectors to retrieve relevant document segments. Searches span both detailed and summarized levels for optimal accuracy.
- Results Presentation: Agents compile search results, removing duplicates and redundant data before presenting the final output.