Campus Co-Pilot

An autonomous AI agent built for TUM (Technical University of Munich) students. Developed at the TUM.ai Makeathon 2026. Combines a FastAPI backend with AWS Bedrock (Claude 3.5 Sonnet) and a React frontend styled as a macOS-inspired desktop environment.

Architecture

Frontend (React + TypeScript)
        ↕  REST / WebSocket
Backend (FastAPI)
        ↕
AWS Bedrock — Claude 3.5 Sonnet (orchestrator)
        ↕
Cognee (SQLite + FastEmbed) — memory and RAG
        ↕
Deepgram (live STT) / ElevenLabs (streaming TTS)

The orchestrator routes each user request to one or more specialized agents:

Agent	Responsibility
Moodle Agent	Retrieves and summarizes lecture slides and course documents
Agenda Agent	Tracks deadlines, exams, and scheduled events
Room Agent	Handles study room reservations

Interfaces

• Text chat — conversational interface with full agent routing
• Voice — real-time voice-to-voice via Deepgram (STT) and ElevenLabs (TTS), with karaoke-style word highlighting

Requirements

• Python 3.11+
• Node.js and npm
• AWS credentials with Bedrock access
• Deepgram API key
• ElevenLabs API key

Setup

Backend

pip install -r requirements.txt

Create a .env file:

AWS_DEFAULT_REGION=us-east-1
BEDROCK_MODEL_ID=eu.anthropic.claude-sonnet-4-5-20250929-v1:0
DEEPGRAM_API_KEY=your_key
ELEVENLABS_API_KEY=your_key
ELEVENLABS_VOICE_ID=your_voice_id

Start the server:

python speech_interface.py

Server runs at http://localhost:8000.

Frontend

cd campus-os
npm install
npm start

Frontend available at http://localhost:3000.