Skip to content

christiannp/Telecom-Agent

Repository files navigation

📡 CovMo™ Telecom Agent

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.

Python FastAPI Streamlit License CovMo Platform

Key Capabilities

  • 🔴 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

System Architecture

                                 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.


Multi-agent Design

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.


Dashboard UI

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.


Quick Start

# 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

See GETTING_STARTED.md for full instructions.


Demo Scenario

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.


Demo Questions

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.


Known Limitations

  • 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

Roadmap

  • 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

🤝 Contributing

This is a proprietary demo project. For questions or collaboration inquiries, please contact the project maintainers.


📄 License

Proprietary — CovMo™ Telecom Intelligence Platform Demo

All rights reserved. This software is provided for demonstration purposes only.

About

AI-powered telecom operational intelligence platform to demonstrate real-time network optimization using multi-agent AI orchestration.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors