All work
AI & MLHackathon build, Praxis 2025
SentinelAI
A security scanner for AI models. One command, five checks, one risk score.
The problem
AI models go live in hospitals, banks and hiring systems, yet almost nobody runs a security check on them first.
What I built
- Scans any Hugging Face model for unsafe serialization, dependency CVEs, risky configs, prompt-injection weaknesses and bias
- Rolls every finding into a single 0–100 risk score mapped to the OWASP LLM Top 10
- CLI for developers plus a web dashboard with GitHub sign-in and HTML and JSON reports
- Shipped in ten hours at Praxis 2025: the CLI on PyPI (pip install sentinel-ai-scanner), the backend on Render, the frontend on Vercel
My part on the team
Designed the risk-scoring framework that rolls every finding into one score, and validated the scanner against 15+ Hugging Face models.
How it works
- 1Point it at a model
- 2Run five security scans
- 3Score and weigh the findings
- 4Publish a report and dashboard