DocumentFlow

AI-Powered Document Management for Legal Pros

I built DocumentFlow as a standalone desktop app with Tauri + React, backed by AWS. The client runs in a native window, while the backend uses Python + FastAPI on Lambda, S3 for storage, MongoDB for data, OAuth2 + MFA for auth, and Docker for local dev.

Tauri + React Client

Wrapped my React UI in Tauri for a fast, native-feeling desktop app on Windows, macOS, and Linux.

AWS Backend & Security

Deployed FastAPI in Lambda, set up S3, IAM, CloudWatch logs, plus OAuth2 & MFA for rock-solid security.

Tech Stack

Core Features I Built

File Parsing:

Parsed uploads, preserved tables/lists, chunked and tagged.

Semantic Graph:

Built a Neo4j graph to map entities and relationships.

Vector Store:

Set up Pinecone for high-speed vector similarity searches.

Agent Swarm:

Orchestrated multiple AI agents to handle parsing, QA, and output.

Swarm Lib Refactor:

Refactored OpenAI's swarm library for production (for example adding async tool call support).

File Editing:

Edit files in real-time with AI (docx, pdf, powerpoint, txt).

How It Works

  1. Upload & Segment – Files are parsed, split, and entities are tagged.
  2. Index & Graph – Segments go to Pinecone; entities link up in Neo4j.
  3. Query & Retrieve – Agents vectorize queries and use metadata search on both detailed & summary indexes to fetch the most relevant chunks.
  4. Compile & Return – Responses are de-duplicated and shown in the UI chat.

Graph Visualization

Graph Visualization

RAG Flow Chart

RAG Flow Chart

Demo

DocumentFlow Demo 1
DocumentFlow Demo 2