Inspiration
Data centers power everything from AI models to cloud applications, but a small infrastructure issue can quickly turn into a major outage if warnings are missed or technicians receive incomplete information. Inspired by NorthMark Compute & Cloud's challenge to keep the world running, we wanted to build an AI-powered system that detects problems early, investigates their causes and coordinates the right people to resolve them.
What It Does
404: Panic Not Found is an autonomous data center incident-response platform that monitors computing infrastructure and coordinates recovery through AI agents.
The application features an interactive 3D data center that displays live infrastructure health, including CPU usage, memory consumption and active incidents.
When an issue occurs, technicians receive an email alert and can call our AI agent or use the built-in chat interface. The agents investigate the issue, identify its root cause, evaluate its severity and assign it to the most qualified available technician based on expertise, seniority and work schedule.
Once assigned, the phone call ends while the system executes an authorized recovery action on the technician's behalf. After verifying recovery through updated infrastructure metrics, the application records the completed incident and emails a detailed resolution report to the administrator.
How We Built It
We used NVIDIA Nemotron and LangGraph to develop three specialized agents responsible for infrastructure investigation, risk assessment and technician coordination.
Our backend uses SpacetimeDB to manage technician accounts, incident assignments, infrastructure status and real-time application updates.
We deployed two Docker compute resources on AWS EC2 and integrated Amazon CloudWatch MCP and a custom Docker MCP server to retrieve real infrastructure metrics and investigate resource issues.
For communication, we integrated Twilio for phone calls, Cartesia for AI-generated speech and Gmail SMTP for incident notifications and recovery reports.
The frontend uses HTML, CSS, JavaScript and Three.js to visualize the data center in 3D and display each agent's actions through a live activity panel.
Challenges We Faced
One major challenge was designing an AI system that performs meaningful actions rather than simply generating recommendations. We addressed this by connecting our agents to real infrastructure measurements and implementing recovery workflows that can be verified through updated metrics.
Another challenge was coordinating technician assignments without conflicts. We designed selection rules that consider technical qualifications, work schedules, availability and experience while preventing technicians from being assigned to multiple incidents simultaneously.
We also had to distinguish real infrastructure data from simulated physical facility measurements, such as water consumption and cooling-system conditions.
Accomplishments That We're Proud Of
We designed an end-to-end incident-response workflow that connects live infrastructure monitoring, multi-agent investigation, intelligent technician assignment and automated recovery.
We're especially proud of combining a phone-based AI interface with an interactive 3D data center, allowing users to communicate naturally with agents while watching their progress in real time.
What We Learned
We learned how to design multi-agent systems that interact with real infrastructure, coordinate decisions across multiple tools and maintain shared operational state through SpacetimeDB.
We also gained a better understanding of data center reliability, infrastructure observability and the importance of verifying AI-generated actions rather than assuming they succeeded.
What's Next
We plan to expand 404: Panic Not Found to support additional compute resources, more advanced infrastructure diagnostics and integrations with physical cooling, electrical and water-monitoring systems.
Our long-term goal is to help data center teams prevent outages, coordinate faster responses and ensure critical operational information is never lost between technician shifts.n-Dev
Built With
- amazon-cloudwatch
- aws-ec2
- aws-iam
- cadvisor
- cloudwatch-mcp
- css
- docker
- gmail-smtp
- html
- javascript
- langgraph
- model-context-protocol
- node.js
- spacetimedb
- three.js
- twilio
- typescript
