Team Duck - Challenge 1 Swisscom - Network Anomaly Detection

Inspiration

We wanted to challenge ourselves by working with graph data in a way that went beyond anomaly detection and tackled root cause analysis in real-time. The Swisscom challenge gave us the perfect opportunity to combine graph machine learning with AI-driven insights.

What it does

Our project, the Anomaly Investigation Workbench, detects anomalies in large temporal graphs and empowers engineers to quickly identify faulty devices, misconfigured components, or unusual patterns. It doesn’t just surface problems—it guides users through an investigation funnel, helping them move from high-level triage to root cause insights.

How we built it

We trained a Temporal Graph Network (TGN) to understand “normal” graph behavior in an unsupervised setting. From there, we built a data pipeline to process outputs into actionable datasets and developed an interactive UI to visualize anomalies. We also integrated AI in two roles: as a natural language controller for the UI and as a Co-Pilot to provide expert-level analysis.

Challenges we ran into

The toughest challenge was finding a suitable temporal graph model that could handle dynamic datasets at scale. Balancing performance with usability also required careful architectural choices.

Accomplishments that we're proud of

We’re proud of building an end-to-end proof of concept in hackathon time, delivering strong performance, clear visualizations, and AI-powered interactions.

What we learned

We deepened our expertise in graph data, anomaly detection, and AI integration, and learned how to separate the scoring engine from the investigation UI for future scalability.