VIPÍK – RAG Based Virtual Assistant

University Hospital Brno (FN Brno) includes three hospitals with over 350 departments and more than 7,500 employees. When staff members need to get something done, they must navigate hundreds of internal documents or consult on average three different people. This is a massive overhead. VIPÍK aims to streamline this process using an AI-assistant orchestration system.

_{Created at Hack jak Brno 2025 hackathon}

What It Does 🤖

VIPÍK is an intelligent multi-agent system that serves as a unified interface for hospital employees to navigate complex administrative tasks:

• 📚 Knowledge Agent - Answers questions using hospital documents (RAG with vector search)
• 🏥 Patient Lookup Agent - Searches FHIR database for patient information
• ✈️ Business Trip Agent - Handles travel request submissions and expense reporting
  🔊 Voice/Text Input: Accepts both text and voice queries from users
• 🧭 Router/Classifier - Orchestrates agents based on user intent and role-based permissions

🔐 Role-Based Access: Different employees see have access to different internal documents and functions (e.g., only authorized staff can access patient data, only certain employees can submit trip requests).

How to request a business trip?

Show matched text, Fill out a form

System Overview 🏗️

Architecture

Agentic Architecture

The system uses a router-based multi-agent architecture:

1. User Query → Application interface (role-aware)
2. Intent Classifier → Analyzes query and user role to determine which agent to invoke
3. Orchestrator Agent → Delegates to specialized agents:
   • Knowledge Search Agent - Retrieves from vector DB (hospital policies, processes)
   • Patient Data Agent - Queries FHIR database for patient records
   • Travel Request Agent - Handles future business trip submissions
   • Travel Expense Agent - Processes expense reports for past trips
4. Response Generation → Agent returns structured Czech response with sources
5. User Interface → Displays answer with citations and suggested actions

Extensibility 🔧

• ➕ Add New Agents: Create new specialized agents by defining intent patterns and system prompts
• 🔄 Add New Workflows: Extend existing agents with multi-step workflows (approval chains, integrations)
• 🏢 On-Premise Deployment: Replace OpenAI API with vLLM running local models for air-gapped environments
• 👥 Role Management: Easily configure new roles and permissions in user configuration

Quick Start 🚀

Prerequisites: Docker, Node.js 18+, OpenAI API key, Python 3.9+ (tested with Python 3.12)

_{(Disclaimer) For local use make sure you have CORS enabled and use browser that supports Web Speech API, typically Chrome.}

# 1. Build and Start InterSystems IRIS Database (with FHIR)
cd FHIR-AI-Hackathon-Kit/Dockerfhir
docker pull intersystemsdc/irishealth-community:latest
docker-compose build
docker-compose up -d
cd ../..
# Wait 1-2 minutes for IRIS to fully initialize
2. Setup Backend
cd backend

python -m venv venv

source venv/bin/activate  # Windows: venv\Scripts\activate

pip install -r requirements.txt

cp .env.example .env
Edit .env and add your OPENAI_API_KEY
3. Ingest Data
cd backend

python ../scripts/ingest_data.py
4. Start Backend
cd backend

uvicorn app:app --reload --host 0.0.0.0 --port 8000
5. Start Frontend (new terminal)
cd frontend

npm ci

npm run dev

✨ Access:

• 🌐 Frontend: http://localhost:3000
• ⚙️ Backend API: http://localhost:8000
• 💾 IRIS Portal: http://localhost:32783/csp/sys/UtilHome.csp (user: _SYSTEM, password: ISCDEMO)

Common Tasks 📋

📄 Add New Documents

1. Place .docx or .xlsx files in raw_data/
2. Run: python scripts/ingest_data.py

🗑️ Clear Database and Re-ingest

1. Delete existing data: python scripts/delete_database.py
2. Re-ingest all documents: python scripts/ingest_data.py

Note: Deleting the database will remove all ingested document chunks and requires re-ingestion.

🔄 Change Embedding Model

1. Edit backend/config.py: Update embedding_model and embedding_dimension
2. Clear database: python scripts/delete_database.py
3. Re-ingest data: python scripts/ingest_data.py

🔄 Restart Services

# Restart IRIS
cd FHIR-AI-Hackathon-Kit/Dockerfhir
docker-compose restart
Restart Backend
cd backend

uvicorn app:app --reload --host 0.0.0.0 --port 8000
Restart Frontend
cd frontend

npm run dev

Additional Documentation 📚

• https://github.com/ElNiborus/HackJakBrno25/blob/main/backend/README.md - Backend structure, configuration, and development guide
• https://github.com/ElNiborus/HackJakBrno25/blob/main/frontend/README.md - Frontend development and deployment
• FHIR-AI-Hackathon-Kit Tutorial - InterSystems IRIS and FHIR guides
• https://github.com/ElNiborus/HackJakBrno25/blob/main/CLAUDE.md - Project context for AI assistants