Plan Bahn

AI-powered Deutsche Bahn travel assistant with real-time data, intelligent routing, and delay predictions.

Quick Start

Docker (Recommended)

git clone https://github.com/larsboes/optimized-bahn-msghackathon && cd optimized-bahn-msghackathon
cp src/backend/.env.example src/backend/.env  # Add your API keys
docker compose up

Project Structure

src/
├── backend/          # FastAPI + LangGraph (Python)
│   ├── domain/       # Business entities
│   ├── use_cases/    # Application logic
│   ├── interfaces/   # Abstract repositories
│   ├── infrastructure/  # Concrete implementations
│   ├── service/      # Legacy service layer
│   └── agent/        # AI agent with tools
├── frontend/         # Vue 3 + Vuetify
└── data-loader/      # GTFS/NeTEx ingestion service

Tech Stack

Component	Technology
Backend	FastAPI, LangGraph, Pydantic
Frontend	Vue 3, Vuetify, Pinia
AI/LLM	Google Gemini, LangGraph ReAct
Database	SQLite (GTFS data), PostgreSQL (planned)
Data	Deutsche Bahn Timetables API, GTFS, NeTEx
DevOps	Docker Compose, uv

Project Status

This project started as a hackathon MVP and is being modernized for production use.

Done

• Docker Compose setup (backend, frontend, data-loader, PostgreSQL, Redis)
• Data loader service with extensible loader architecture
• Clean Architecture folder structure (domain, use_cases, interfaces, infrastructure)
• Real-time integration with DB Timetables API
• LangGraph AI agent with tool calling
• Vue 3 frontend with chat interface
• GTFS data ingestion pipeline
• Network graph for pathfinding suggestions

In Progress

• Migrate from SQLite to PostgreSQL + PostGIS
• Redis session management (currently in-memory)
• Full station search in AI agent (currently ~10 hardcoded stations)
• Global error handling and structured logging

Planned

• Refactor Docs: Reorganize and expand documentation for clarity
• Backend Refactoring: Clean up legacy code and improve type safety
• Frontend Refactoring: Consolidate components and improve performance
• React + Bun Migration: Migrate from Vue 3 + npm to React + Bun
• Advanced Routing: CSA (Connection Scan Algorithm) or Raptor for multi-hop connections
• Historical Delays: ML predictions using piebro/deutsche-bahn-data
• Platform-aware Transfers: Use NeTEx data for realistic transfer times
• HashiCorp Vault: Enterprise secrets management

Known Limitations

1. AI Agent Station Limit: Tool can only look up ~10 hardcoded stations (see src/backend/agent/tools.py)
2. Routing: No automatic multi-hop routing (A→B→C→D); requires explicit "via" station
3. Simulated Data: Wagon loads and some delays are randomized, not from real API
4. Session Storage: In-memory only, doesn't persist across restarts

Documentation

For comprehensive documentation on architecture, setup, usage, and API reference, see docs/README.md.

Contributing

We welcome contributions! Please see CONTRIBUTING.md for:

• Getting started with development
• Code style and commit conventions
• Testing requirements
• Areas where help is needed
• Bug reporting

License

See LICENSE file.