By: Ayush, Joe, Max, Sophia
Submission Link: https://devpost.com/software/the-most-optimal-network
Every minute of network downtime costs enterprises thousands of dollars — yet most monitoring systems are reactive, not preventive.
We wanted to change that by giving users — from everyday households to enterprise IT teams — a live, intelligent pulse on their network performance.
With the help of the T-Mobile 5G Network, a Raspberry Pi, and an Arduino hardware kit, we set out to build NetPulse, an embedded AI system that brings predictive network diagnostics to the edge.
NetPulse is an end-to-end, real-time intelligent system that:
🔹 Collects environmental, motion, and system metrics from onboard sensors (temperature, humidity, CPU load, ping latency, packet loss, etc.)
🔹 Processes and classifies this data through an edge anomaly detection model running locally on the Raspberry Pi
🔹 Streams all metrics and AI insights to a mobile dashboard via Firestore and an MCP Server, providing instant visualization and personalized optimization tips
📱 Mobile App Features:
- Live graphs of network health metrics (latency, stability, Wi-Fi signal, etc.)
- AI-powered diagnostics & recommendations (e.g., “Reduce router distance” or “High packet loss detected — potential interference”)
- Realtime data sync between Arduino → Phone → MCP Server → Firestore
Hardware: Arduino + environmental & IMU sensors to capture physical and performance parameters
Networking: T-Mobile 5G backbone for stable and fast data transmission
Backend: Python-based Firestore pipeline for structured logging and real-time analytics
AI Layer: Anomaly detection model trained to identify irregular system behavior and generate contextual diagnostics — integrated with an MCP Server running Claude Sonnet 4.5
Frontend: Mobile app interface for live metric visualization and AI feedback
- Building a robust data pipeline from Arduino → Mobile → MCP → Firestore with low latency
- Managing real-time synchronization between Arduino and Raspberry Pi sensors
- Ensuring consistent calibration and normalization for ML inference
- Designing an ML model that performs well on small, noisy, and unlabeled datasets
- End-to-end integration across hardware, AI, and cloud — with a fully functional live demo
- Successfully collected and parsed thousands of real-world sensor readings
- Built a responsive mobile UI that visualizes real-time network health
- Developed a modular ML inference architecture for seamless edge-to-cloud deployment
- How system-level metrics like packet loss, CPU load, and Wi-Fi signal strength can predict instability before users notice
- The importance of cleaning and scaling sensor data for reliable ML predictions
- How to design for resilience — ensuring fault-tolerant data flow from embedded to cloud