Sentinel-AI
A multi-agent cloud-security incident-response platform built during the NVIDIA x ASUS Hack-a-Claw hackathon.
Role
Full-stack / AI systems contributor
Team
Hackathon team — NVIDIA x ASUS Hack-a-Claw
Status
Completed
Timeline
May 2026
- ▹Built in under 24 hours
- ▹Incident summaries in under 30 seconds
- ▹Multi-agent orchestration
- ▹Next.js + FastAPI + Docker + Supabase
- ▹GPU-hosted AI inference
Problem Statement
Cloud security incidents produce noisy, high-volume telemetry. Responders need fast, structured summaries that classify threats and propose next actions.
Design Goals
- Multi-agent workflow for classification, analysis, and reporting
- Incident summaries in under 30 seconds
- Deployable full stack within a 24-hour hackathon window
System Architecture
Next.js frontend → FastAPI orchestration → agent workers using NVIDIA Nemotron on GPU inference → Supabase for persistence, packaged with Docker.
Software Design
Autonomous agents handle threat classification, cloud event analysis, and report generation. The API coordinates distributed workflows and returns condensed incident summaries to the UI.
Results
Shipped a working platform during the competition that generated AI incident summaries in under 30 seconds, built and deployed in under 24 hours.
Challenges
- Integrating multiple services under extreme time pressure
- Keeping agent outputs grounded and structured
- Coordinating GPU inference with the web stack
Lessons Learned
Clear agent contracts and a thin orchestration layer beat a monolithic prompt when time is limited.
Future Improvements
- Stronger evaluation harness for summary quality
- Richer connector set for cloud event sources
Questions about this project? Email Caleb.