Milan AI Week 2026 · AI Agent Olympics
Tracks: Intelligent Reasoning · Enterprise Agent · Speechmatics Sponsor Prize
Q-Optima is a multi-cloud autonomous AI agent that solves NP-Hard logistics optimization problems using Quantum Computing. A logistics commander speaks a voice instruction, uploads a route map, and within 30 seconds receives a quantum-optimized delivery route — enriched with real-time weather, traffic, and carbon data — delivered to their phone via Telegram.
🎙 Speak constraint → 📸 Upload map → ⚛️ Quantum solve → 📍 Optimal route → 📱 Dispatch alert
┌─────────────────┐ ┌──────────────────────────────────────────┐ ┌──────────────────┐
│ Vercel │────▶│ Render — FastAPI + LangGraph (7 nodes) │────▶│ IBM Quantum │
│ Next.js 14 │ │ │ │ QAOA / Aer │
│ Leaflet map │ │ ┌──────────┐ ┌──────────┐ │ └──────────────────┘
│ WebSocket UI │ │ │Speechmatics│ │Gemini 2.5│ │
└─────────────────┘ │ │ STT │ │ Vision │ │
│ └──────────┘ └──────────┘ │
│ ┌──────────┐ ┌──────────┐ │
│ │Open-Meteo│ │ TomTom │ │
│ │ Weather │ │ Traffic │ │
│ └──────────┘ └──────────┘ │
│ ┌──────────┐ ┌──────────┐ │
│ │ Climatiq │ │ Telegram │ │
│ │ Carbon │ │ Dispatch │ │
│ └──────────┘ └──────────┘ │
└──────────────────────────────────────────┘
│
┌─────────────▼────────────┐
│ AMD Dev Cloud │
│ Whisper ROCm (GPU STT) │
│ via Cloudflare Tunnel │
└──────────────────────────┘
Speechmatics STT → Gemini Vision → Enrich (OSRM+Weather+Traffic+Carbon)
→ Build QAOA Circuit → Execute IBM Quantum → Self-Reflect → Parse+Finalize
↑ │
└─────── retry with p+1 if quality fails ┘
Node 6 — Self-reflection loop: The agent autonomously checks if the quantum result is a valid permutation and if confidence exceeds the threshold. If not, it increases QAOA layers and retries — no human intervention.
| Layer | Technology |
|---|---|
| Frontend | Next.js 14 (App Router), Tailwind, Framer Motion, Leaflet.js |
| Orchestrator | FastAPI, LangGraph, Python 3.11 |
| Primary STT | Speechmatics batch + RT transcription API |
| Vision AI | Google Gemini 2.5 Flash Lite (multimodal) |
| Quantum | IBM Quantum / Qiskit QAOA, Aer simulator |
| AMD GPU | OpenAI Whisper on ROCm PyTorch (perception node) |
| Road data | OSRM public API (real driving distances) |
| Weather | Open-Meteo API (zero cost, no key) |
| Traffic | TomTom Flow Segment Data API |
| Carbon/ESG | Climatiq freight emission API |
| Dispatch | Telegram Bot API |
| Analytics | Supabase PostgreSQL (job history) |
| Hosting | Vercel (frontend) + Render (backend) + AMD Dev Cloud |
| Tunneling | Cloudflare Tunnel (free HTTPS) |
Total cost: $0.00 out of pocket.
The quantum distance matrix is not just geographic — it encodes real-world constraints:
Enriched distance[i][j] =
OSRM road distance (km)
+ weather penalty (Open-Meteo WMO severity × max_penalty)
+ traffic penalty (TomTom congestion ratio × max_penalty)
+ carbon penalty (Climatiq CO₂e × cost_weight)
QAOA minimizes this enriched matrix — simultaneously optimizing for distance, weather safety, congestion, and carbon footprint in a single quantum pass.
q-optima/
├── frontend/ # Next.js 14 — Vercel
│ ├── app/
│ │ ├── globals.css
│ │ ├── layout.tsx
│ │ ├── page.tsx
│ │ └── components/
│ │ ├── QuantumCommandCenter.tsx
│ │ ├── AgentTimeline.tsx
│ │ └── VoiceOutput.tsx
│ └── package.json
├── backend_vultr/ # FastAPI + LangGraph — Render
│ ├── main.py
│ ├── agent/
│ │ ├── graph.py # 7-node StateGraph with reflection loop
│ │ ├── nodes/
│ │ │ ├── speechmatics_node.py
│ │ │ ├── gemini_node.py
│ │ │ ├── qiskit_node.py
│ │ │ ├── ibm_quantum_node.py
│ │ │ └── reflection_node.py
│ │ └── tools/
│ │ ├── osrm_tool.py
│ │ ├── weather_tool.py
│ │ ├── traffic_tool.py
│ │ ├── carbon_tool.py
│ │ ├── telegram_tool.py
│ │ └── supabase_tool.py
│ ├── models/schemas.py
│ └── utils/
│ ├── qubo_parser.py
│ └── classical_comparison.py
├── backend_amd/ # Whisper ROCm — AMD Dev Cloud
│ ├── whisper_service.py
│ └── Dockerfile
└── docker-compose.yml
- Python 3.11
- Node.js 18+
- API keys (see
.env.example)
# Clone
git clone https://github.com/YOUR_USERNAME/q-optima.git
cd q-optima
# Backend
cd backend_vultr
cp .env.example .env # fill in your API keys
pip install -r requirements_vultr.txt
uvicorn main:app --reload --port 8000
# Frontend (new terminal)
cd frontend
npm install
echo "NEXT_PUBLIC_BACKEND_URL=http://localhost:8000" > .env.local
npm run devOpen http://localhost:3000 → click Run Quantum Agent.
| Key | Source | Cost |
|---|---|---|
GEMINI_API_KEY |
aistudio.google.com | Free |
SPEECHMATICS_API_KEY |
portal.speechmatics.com | $200 credit |
IBM_QUANTUM_TOKEN |
quantum.ibm.com | Free |
TOMTOM_API_KEY |
developer.tomtom.com | Free 2500/day |
CLIMATIQ_API_KEY |
climatiq.io | Free 1000/mo |
TELEGRAM_BOT_TOKEN |
@BotFather | Free |
SUPABASE_URL + SUPABASE_ANON_KEY |
supabase.com | Free |
No-key APIs: Open-Meteo, OSRM, Leaflet/OSM, Web Speech API.
render.com → New Web Service → Connect GitHub
Root Directory: backend_vultr
Build: pip install -r requirements_vultr.txt
Start: uvicorn main:app --host 0.0.0.0 --port $PORT
Add all env vars from .env.example
vercel.com → Import repo
Root Directory: frontend
NEXT_PUBLIC_BACKEND_URL = https://your-render-url.onrender.com
| Criterion | How Q-Optima addresses it |
|---|---|
| Originality | First agent combining multimodal LLM + quantum computing for real enterprise logistics |
| Application of Technology | 14 real APIs, Speechmatics as primary STT, IBM QAOA, LangGraph self-reflection |
| Business Value | Carbon scoring, classical vs quantum comparison %, Telegram dispatch, Supabase analytics |
| Presentation | Live Vercel URL, animated agent timeline, voice output, Telegram demo moment |
MIT — built for Milan AI Week 2026 AI Agent Olympics.