Intent-Based RAN Optimization · Urban Mobility Intelligence · AI Autonomous Operations
CovMo™ is an enterprise-grade telecom intelligence platform that simulates the Taipei Arena Power Station Concert Egress scenario (May 15, 2026, 22:00). The system demonstrates how AI can autonomously optimize telecom networks during mass egress events.
- 🔴 Real-time Level-2 RAN Telemetry — RSRP, SINR, TA, PRB, CQI streaming at 500ms intervals
- 🤖 Multi-Agent AI Orchestration — 5 specialist agents coordinated via Google ADK
- 🗺️ Urban Mobility Digital Twin — Taipei Arena → Nanjing Fuxing MRT crowd simulation
- ⚡ Autonomous Optimization — Policy-validated AI actions with 85%+ confidence threshold
- 📊 Subscriber-Level QoE Analytics — VIP tracking with frustration index & degradation prediction
- 🌧️ Weather-Aware Intelligence — Rainfall impact on mobility patterns (7.2mm/hr scenario)
- 📈 Executive KPI Dashboard — Real-time business metrics with Plotly + Folium visualization
- 🔁 Continuous Monitoring Loop — Persistent alert escalation with escalating intervention levels
- 📽️ Incident Replay — Snapshot-based historical scrubbing at major incident boundaries
- 🔗 Correlated Event Pipeline — 6 unified scenario detectors cross-correlating RAN + Mobility + Context signals
Streamlit Dashboard (8500)
↓
FastAPI Server (Port 8400)
↓
Telemetry Streamer
↓
AI Correlation & Analytics Layer
↓
Google ADK Multi-Agent Orchestration
┌───────┬──────────┼──────────┬─────────┐
│ RAN │ Mobility │ Context │ Policy │
│ Agent │ Agent │ Agent │ Agent │
└───────┴──────────┴──────────┴─────────┘
See ARCHITECTURE.md for full details.
The system uses 5 specialist agents coordinated by a root Intent Orchestration Agent via Google ADK:
- RAN Agent — Radio telemetry analysis: RSRP, SINR, PRB, CQI, handover failure prediction, signal cliff & anomaly burst detection
- Mobility Agent — Crowd dynamics: MRT congestion (GREEN/YELLOW/RED), YouBike availability, egress velocity, slip risk
- Context Agent — Environmental awareness: Taiwan CWA weather integration, walking propensity, rainfall impact on mobility
- Policy Agent — Autonomous governance: validates all actions (≥85% confidence, ≥10% KPI improvement), VIP SLA enforcement, loop detection
- Intent Orchestrator — Routes user queries to the right agent(s), explains AI reasoning with confidence scores
See AGENTS.md for full details.
The dashboard combines real-time telemetry, mobility visualization, and autonomous action control in a single Streamlit interface:
- Executive KPI Panel — 8 live metrics: Subscriber Satisfaction, VIP QoE, Congestion Risk, AI Confidence, SLA Health, Revenue Protection, Mobility Pressure, and Escalation Level (1–4)
- Live Telemetry Charts — RSRP/SINR/PRB trends, Timing Advance mass-egress indicator, handover success rate, and PRB congestion heatmap
- Mobility Digital Twin — Folium map with Taipei Arena + MRT markers, subscriber dots (color-coded by signal quality, VIPs enlarged), cell sector overlays, and YouBike station
- AI Reasoning Console — Real-time chain-of-thought display per agent with confidence scores and policy decisions
- Autonomous Actions — Manually trigger or review approved/blocked actions (VIP Priority Routing, Load Balancing, Micro-cell Handover, etc.) with reasoning and expected KPI impact
See DASHBOARD.md for full details.
# 1. Install dependencies
pip install -r requirements.txt
# 2. Configure .env file
cp .env.example .env
# Add your OLLAMA_API_KEY to .env
# 3. Run auto script
./run.sh- Dashboard: http://localhost:8500
- ADK Agents: http://localhost:8080
- API: http://localhost:8400
See GETTING_STARTED.md for full instructions.
The Taipei Arena Power Station Concert Egress (May 15, 2026, 22:00) simulates ~1,500 subscribers exiting a concert under heavy rain (0→12 mm/hr). The crowd funnels toward the Nanjing Fuxing MRT, creating a cascading network crisis across 7 incident arcs:
| Arc | Trigger | Autonomous Response |
|---|---|---|
| VIP Degradation | VIP RSRP < −105 dBm underground | VIP Priority Routing approved at 92% confidence |
| MRT Overload Cascade | MRT DAS cell PRB > 90% | Temporary Load Balancing triggered |
| Weather Transition | Rain spikes 0→12 mm/hr at tick ~50 | MRT capacity reallocation + slip-risk alert |
| Handover Storm | Underground transition phase 0.65–0.80 | Micro-cell Handover + DAS steering |
| Anomaly Burst | 20% UEs with CQI 2–3 for 10 ticks | get_anomaly_report() + diagnostic scan |
| YouBike Starvation | All 60 docks empty | Frustration index + MRT pressure alert |
| Secondary Congestion | Neighboring cell PRB > 85% from load-balance | Action blocked, alternative path proposed |
All 7 arcs hit within a 20-minute window, stress-testing the multi-agent system under realistic cascading conditions — weather shifts, underground signal decay, VIP SLA breaches, and policy loop conflicts.
See SCENARIO.md for full timeline and trigger details.
Users interact with the multi-agent system via natural language — queries are routed to the appropriate specialist agent (RAN, Mobility, Context, or Policy) and resolved autonomously. All autonomous actions require ≥85% confidence and ≥10% expected improvement before Policy agent approval. Example queries include:
| Domain | Example Questions |
|---|---|
| Orchestrator | "What is happening right now in the network?", "Should I be concerned in the next 5 minutes?" |
| RAN | "Show me active signal cliffs and handover failure rates", "Generate a full anomaly report" |
| Mobility | "What is the MRT congestion status at each exit?", "What is the current slip risk?" |
| Context | "How is weather affecting subscriber behavior?", "Calculate walking propensity" |
| Policy | "Should I enable VIP Priority Routing now?", "Would load balancing cause secondary congestion?" |
| VIP Analytics | "Show top 10 VIP subscribers by QoE degradation", "Predict SLA breach in the next 5 minutes" |
See USER_GUIDE.md for the full question library.
- LLM Calls: Agents defined but not actively called (requires Ollama API)
- Weather API: Uses mock data (Taiwan CWA integration ready but not active)
- YouBike API: Uses mock data (real API integration ready)
- Historical Replay: Not yet implemented
- Multi-cell Handover: Simplified model
- Real Taiwan CWA API integration
- Real YouBike API integration
- Historical replay mode with timeline scrubbing
- Multi-cell handover visualization
- SON (Self-Organizing Network) optimization loop
- Subscriber journey replay
- Executive PDF report generation
- Prometheus metrics export
- Grafana dashboard integration
- Docker containerization
- Kubernetes deployment manifests
This is a proprietary demo project. For questions or collaboration inquiries, please contact the project maintainers.
Proprietary — CovMo™ Telecom Intelligence Platform Demo
All rights reserved. This software is provided for demonstration purposes only.