RAG HR Assistant

A minimal RAG (Retrieval-Augmented Generation) system with a Flask backend, React frontend, and Qdrant vector database — all containerized with Docker Compose.

Architecture

rag-ui (React + nginx)  :5173
   ↓ API calls
bot (Flask)             :5000
   ↓ vector search
qdrant (vector DB)      :6333

Prerequisites

• Docker & Docker Compose
• API keys for Voyage AI and Groq

Setup

1. Add your API keys

Create bot/.env:

VOYAGE_API_KEY=your-voyage-api-key
GROQ_API_KEY=your-groq-api-key

2. Add your data

The bot/data/ directory is git-ignored for privacy. You need to provide it yourself.

Expected structure:

bot/data/
  └── chunks/
      ├── chunking.json           # chunked text from your PDF
      ├── embeddings.json         # pre-computed Voyage embeddings
      ├── discarded_chunks.json
      └── gold_output.json

The embeddings.json file is loaded into Qdrant automatically on first startup. Without it, the bot has no data to search.

3. Run

docker compose up --build -d

• UI: http://localhost:5173
• API: http://localhost:5000/ask
• Qdrant dashboard: http://localhost:6333/dashboard

4. Stop

docker compose down

Qdrant data persists in a Docker named volume (qdrant_data). To wipe it: docker volume rm rag_qdrant_data.

Running without Docker

# Backend
cd bot
python -m venv bot-env && source bot-env/bin/activate
pip install -r requirements.txt
python src/main.py            # requires Qdrant running on localhost:6333

# Frontend
cd rag-ui
npm install && npm run dev