Overview
CenterGuard helps data center teams prevent costly outcomes by combining real-time monitoring, portable sensor verification, and AI-powered simulations. Its dashboard helps operators find potential problems, compare built-in sensor alerts against independent readings, and assess how failures could spread before deciding what action to take. CenterGuard also coordinates workers with assigned tasks and checklists, all while time-stamped logs and automated log reviews help trace changes, preventing repeated mistakes.
Inspiration
Data center works face a difficult decision when an alert appears, whether to act immediately or investigate further while damage could spread. We created CenterGuard to simplify that decision making process by helping operators understand what is happening, what could happen next, and which response could help most.
What it does
CenterGuard combines an interactive dashboard, a portable sensor prototype, and AI-assisted simulations. Operators can investigate warnings, compare installed sensor readings with independent measurements, and look at potential responses. Assigned tasks, checklists, and timestamped logs help teams coordinate their work and preserve information between shifts, allowing for seamless workflow between workers.
How we built it
Hardware: The prototype was built using Arduino UNO R4 with various sensors available to us. It also included readings displayed on 2 OLED Screens.
Software: The dashboard was building using React, TypeScript, and Three.js. The simulation was built from combining a physics-based model (for heat and water flow) with around 2000 statistically simulated trials for each response. NVIDIA Nemotron then proposes additional points of failure and explains the engines results. Nemotron's work was intentionally separated from calculations to prevent hallucinations from impacting the mathematical data.
Challenges
The main challenge was to incorporate AI with it hallucinating results. To overcome this, we separated numerical modeling from AI work, where Nemotron was used to support the simulation by elaborating on simulated data. Simplifying dashboard UI to be intuitive was another challenge we ran into.
What's Next
Currently, a prototype for the portable sensor is built, however many wires are unprotected and it may not survive in a work environment. While we do have a version two modeled that is much more compact and protected, we hope to test it in a working environment. Additionally, the current simulation time is longer than desired, given the multiple layers and setups in simulation, so a next step would be reduce the time for faster decision making.
Built With
- arduino
- nemotron
- react
- three.js
- typescript