Sentra – RAG for Wissenschaftliche Dienste

A Retrieval-Augmented Generation (RAG) prototype that enables semantic search and question-answering over documents from the Wissenschaftliche Dienste des Deutschen Bundestages.

Architecture

┌──────────┐      ┌──────────┐      ┌──────────┐
│ Frontend │─────▶│ Backend  │─────▶│  Qdrant  │
│ React    │ :3000│ FastAPI  │ :8000│ VectorDB │ :6333
└──────────┘      └──────────┘      └──────────┘
                        │
                   AI Hub API
                  (Embeddings +
                   Generation)

• Frontend — React + Vite + Tailwind + shadcn/ui
• Backend — FastAPI + Docling (PDF parsing) + OpenAI-compatible AI Hub
• Vector DB — Qdrant (cosine similarity, 4096-dim embeddings)
• Models — Octen-Embedding-8B (embeddings), llama-3-3-70b (generation)

Prerequisites

• Docker & Docker Compose (for Docker setup)
• Node.js >= 20 and Python >= 3.12 with uv (for local dev)
• AI Hub credentials (base URL + API key)

---

Option 1: Docker (recommended)

The simplest way to run the full stack.

1. Configure environment

cp 02_backend/.env.example .env

Edit .env and set your AI Hub credentials:

AI_HUB_BASE_URL=https://your-hub-url.example.com/v1
AI_HUB_API_KEY=your-virtual-key-here

The rest of the defaults work as-is for Docker.

2. Start all services

docker compose up --build

This starts three services:

Service	URL	Description
Frontend	http://localhost:3000	Search UI
Backend	http://localhost:8000	FastAPI + Swagger docs at /docs
Qdrant	http://localhost:6333	Vector DB dashboard

3. Ingest documents

Open the frontend at http://localhost:3000, navigate to Dokumente, and click Dokumente einlesen. This parses all PDFs in 03_data/Ausarbeitungen/ and indexes them into Qdrant.

Alternatively, via API:

curl -X POST http://localhost:8000/api/ingest

4. Search

Go to the Suche tab and ask a question in German, e.g.:

Welche Regelungen gelten für die Immunität von Abgeordneten?

Stop

docker compose down

To also clear the vector database:

docker compose down -v

---

Option 2: Local Development (without Docker)

Run each component separately for faster iteration.

1. Start Qdrant

You still need Qdrant running. The easiest way is via Docker:

docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant:latest

2. Backend

cd 02_backend

# Create and configure environment
cp .env.example .env
# Edit .env — set AI_HUB_BASE_URL, AI_HUB_API_KEY
# Set DOCUMENTS_DIR to the absolute path of your PDFs:
#   DOCUMENTS_DIR=/absolute/path/to/03_data/Ausarbeitungen

# Install dependencies
uv sync

# Run the server
uv run uvicorn sentra.main:app --reload --host 0.0.0.0 --port 8000

The backend is now at http://localhost:8000 (Swagger UI at http://localhost:8000/docs).

3. Frontend

cd 01_frontend

# Install dependencies
npm install

# Run dev server
npm run dev

The frontend is now at http://localhost:3000 and proxies API calls to the backend at localhost:8000.

4. Ingest & search

Same as Docker — navigate to Dokumente → Dokumente einlesen, then switch to Suche.

---

Project Structure

├── 00_aisc/              # Branding assets (HPI/AISC logos)
├── 01_frontend/          # React frontend
│   ├── src/
│   │   ├── components/   # UI components
│   │   ├── lib/          # API client, utilities
│   │   └── types/        # TypeScript interfaces
│   └── Dockerfile
├── 02_backend/           # FastAPI backend
│   ├── src/sentra/
│   │   ├── api/          # Routes, request/response models
│   │   ├── ingestion/    # PDF parsing, metadata extraction, chunking
│   │   ├── rag/          # Embeddings, vector store, answer generation
│   │   └── services/     # Ingestion and query orchestration
│   └── Dockerfile
├── 03_data/              # Sample PDF documents
│   └── Ausarbeitungen/
└── docker-compose.yml

API Endpoints

Method	Endpoint	Description
POST	/api/query	Ask a question (supports SSE streaming)
POST	/api/ingest	Parse and index all PDFs
GET	/api/documents	List indexed documents
GET	/api/health	Health check + Qdrant status

---

Acknowledgements

The AI Service Centre Berlin Brandenburg is funded by the Federal Ministry of Research, Technology and Space under the funding code 01IS22092.