DocAtlas

DocAtlas is a multi-tenant platform that helps healthcare organizations turn internal documents into a Patient Navigation Assistant experience.

Built for Netcompany-hackathon presentation: clear problem, practical architecture, and production-oriented implementation.

The Problem

Hospitals and organizations store critical knowledge in scattered files (PDFs, DOCX, spreadsheets), making it hard for staff to quickly retrieve reliable answers.

Our Solution

DocAtlas provides:

• An Admin Portal to securely manage tenant-scoped knowledge files.
• An AI Retrieval Backend (RAG) to process, chunk, and accurately query documents.
• An Embeddable Chat Widget for external websites, securely locked to the tenant's identity via strict origin validation.

Architecture Diagram

Technologies Used

• React + TypeScript + Vite
• Tailwind CSS + ShadCN
• Firebase Auth, Firestore, Storage
• FastAPI (Python)
• ChromaDB Cloud
• Hybrid Retrieval Pipeline (Dense + Sparse embeddings via RRF)
• File parsing pipeline (PDF, DOCX, CSV, XLSX, TXT)
• OpenAI LLM Integration

Run our app

1. Clone our project using:

   git clone https://github.com/Retsos/DocAtlas.git

2. Run these to create the .env files:

cp frontend/DocAtlas/.env.template frontend/DocAtlas/.env
cp backend/.env.template backend/.env

3. Run with Docker:

docker compose up -d

4. Retrieve the test archives from the test_data folder.

This starts:

• backend on 8000
• frontend/DocAtlas served on 5173
• frontend/chatbotClient on 5174