SentinelAI

## Inspiration

Security analysts can't keep up with thousands of network events per second. We built an AI co-pilot that detects
threats, classifies them, and tells you what to do about it.

## What it does

SentinelAI analyzes network traffic and catches the bad stuff:

• Detects anomalies using machine learning
• Classifies threats (DDoS, brute force, malware, data exfiltration, port scans)
• Scores risk so you know what to fix first
• Maps attacks to MITRE ATT&CK framework
• Suggests specific remediation steps
• Visualizes attacker → target connections in real-time

## How we built it

Frontend: Next.js + TypeScript + Tailwind + Recharts Backend: Python FastAPI + scikit-learn ML: Isolation Forest (anomaly detection) + Random Forest (threat classification) Data: 1,000+ synthetic network logs with realistic attack patterns

We trained two models, built an enrichment pipeline that adds risk scores and MITRE techniques, then wrapped it in a dashboard that doesn't suck.

## Challenges

• scikit-learn version conflicts between training and deployment
• Making ML inference fast enough for 100+ events per page load
• Balancing risk score formulas to be actually useful
• Building a professional UI on a hackathon deadline

## Accomplishments

✅ Real trained ML models (not fake demos) ✅ MITRE ATT&CK integration like production security tools ✅ Network topology visualization that's actually cool ✅ Production-quality dashboard with glassmorphism and smooth animations

## What we learned

ML deployment ≠ ML training. Versioning, caching, and performance matter. Cybersecurity domain knowledge runs deep (MITRE, threat taxonomies, kill chains). Real-time visualization is way harder than it looks.

## What's next

• Connect to real network logs (Zeek, Suricata) instead of synthetic data
• Automated response via firewall/SIEM APIs
• Deep learning models for sequential attack patterns
• SaaS platform for actual SOC teams

Built With

• css
• cybersecurity
• fastapi
• groq
• mitre
• next.js
• python
• recharts
• scikit-learn
• tailwind
• typescript