Main Dashboard (Cluster List View)
Main Dashboard (Cluster Location Map View)
Main Dashboard (Cluster List View) w/ AI Assistant
3D Cluster Digital Twin and Airflow Visualization

ThermaMind - AI-Powered Data Center Optimization

We’ve always heard about how AI utilization and data centers contribute heavily to environmental damage, but we wanted to understand why and see if there was a better way forward. Through our research, we discovered a striking fact: 40–60% of the energy footprint in modern AI infrastructure isn’t compute at all — it’s cooling and idling overhead. That insight became the foundation of ThermaMind.

ThermaMind at its core is a real-time GPU cluster monitoring and optimization platform with an integrated AI-powered sustainability assistant that monitors real-time telemetry from simulated GPU clusters, detects cooling inefficiencies, and generates actionable insights and audio alerts to help operators reduce waste and CO₂ emissions.

We built it with a real-time Node.js WebSocket backend, Gemini AI for analysis, ElevenLabs TTS for accessibility, and a React + Tailwind dashboard for visual clarity. The assistant can communicate insights through both text and speech, with mute/unmute support, ensuring accessible interaction in any environment.

Our biggest challenges were:

Understanding the full scope of the problem posed by the shift to more power intensive AI infrastructure, and building an effective solution that addresses it.
Building a reliable real-time telemetry stream while maintaining low-latency.
Synchronizing AI generation and voice output
Making the experience intuitive and scalable under hackathon constraints.

In the end, we created a scalable, impactful, and accessible solution that tackles a real sustainability problem head-on: unnecessary cooling waste in data centers.

ThermaMind directly aligns with ConocoPhillips’ vision of AI-driven solutions for real-world sustainability challenges, demonstrating:

Accessibility: voice interaction, simple UX
Impact: tackling energy waste & CO₂ reduction
Technical depth: real-time streaming, AI analysis, voice synthesis
Scalability: can integrate into real data centers and expand beyond single GPU clusters.

Built With

Submitted to

CodeRED Astra
- Winner Track 2: ConocoPhillips Track

Created by

I worked on the product design + frontend implementation for ThermaMind. Integrating most of the backend features with the frontend. Some of the features I worked on include: GPU cluster status (heatmap), GPU clusters list, key dashboard statistics, and the real-time energy & workload chart, 3D airflow visualization for the GPU clusters, etc.

Joshua Eworo
I primarily focused on the backend infrastructure and global map integration — ensuring that ThermaMind could deliver live, accurate, and visually dynamic telemetry data across multiple data centers. My work bridged the real-time Node.js backend with the React front-end, enabling low-latency communication and clear visualization of global GPU activity.

My key responsibilities included:

Backend Framework: Set up the Node.js + Express backend and WebSocket communication layer to stream telemetry data in real time between the simulated GPU clusters and the dashboard.

Map Integration: Implemented the Mapbox API to visualize global cluster locations, configuring live markers, dynamic color-coding, and region-specific telemetry updates.

Data Synchronization: Linked backend telemetry with the frontend map and dashboard components, ensuring seamless updates without page reloads.

Optimization & Testing: Debugged backend endpoints and refined update intervals to maintain low latency, smooth map performance, and accurate cooling and GPU load readings.

In short, I built the real-time foundation and interactive mapping layer that connected ThermaMind’s backend intelligence with its user-facing experience — making the platform both technically robust and visually intuitive.

DavidAOdim Odim
I focused on building the real-time AI and voice assistant layer of ThermaMind. I integrated the Gemini model with live telemetry through WebSockets, implemented a telemetry buffer for efficient data streaming, and helped stabilize backend services.
I also led the voice accessibility feature using ElevenLabs TTS, adding mute/unmute controls and ensuring smooth audio playback on the frontend. My work bridged real-time data with AI insights, making the assistant both intelligent and accessible, while supporting the project’s scalability and sustainability goals.

Hugo Santos