LIDC Computing Department · Digital Future Fest 2026
Catch real hackers in the act. Let AI explain every attack.
SENTINEL places decoy pages on the open internet, detects attacks with signature rules and machine learning, blocks repeat attackers automatically and streams every event to a live security dashboard.
Free to run · No real systems attacked · Real and simulated traffic always labelled
How it works
From the first probe to a blocked attacker in about a second
- Step 1
Trap
Decoy pages such as /wp-login.php, /.env and /phpmyadmin sit on the open internet and attract real bots and scanners.
- Step 2
Detect
Every request is checked by signature rules and a machine-learning model. When they disagree, the AI can still catch what rules miss.
- Step 3
Respond
Critical or repeated attacks get the IP blocked for 24 hours, and the event reaches the dashboard in about a second.
Modules
A small security operations center, built from free tools
Web honeypot
Fake WordPress, phpMyAdmin, admin logins, leaked .env and .git files. Anything that touches them is logged and analysed.
Hybrid AI detection
Suricata-style signatures plus a TF-IDF and logistic regression model trained in Python and run in TypeScript on the edge.
Automatic response
A SOAR-style playbook blocks attackers after one critical hit or five malicious requests in ten minutes.
Live SOC dashboard
3D attack globe, live feed, MITRE ATT&CK mapping, top attackers and an AI-written incident report for every event.
Red Team console
Visitors launch SQL injection, XSS, brute force and seven more attacks from their phone, or write their own payload.
Privacy by design
Visitor IPs are partly masked, typed passwords are redacted before storage, and simulated attackers use reserved test IPs.
FAQ
Questions judges usually ask
Is SENTINEL attacking anyone?
No. SENTINEL only listens. Attacks from the Red Team console are sent to SENTINEL itself, and simulated attackers use reserved test IP ranges that belong to nobody.
Are the attacks on the dashboard real?
Both kinds appear, and they are always labelled. Real means traffic from the internet that hit a honeypot trap. Sim means it was launched from the Red Team console.
How does the AI work?
A character n-gram TF-IDF model with logistic regression is trained in Python, exported to JSON and executed in pure TypeScript. Its verdict is combined with signature rules and behaviour analysis.
How accurate is it?
Over 99% on attack templates the model never saw in training. That data is synthetic, so real-world accuracy is lower, which is why real honeypot traffic is fed back into training.
What does it cost to run?
Nothing. It runs on the free tiers of Vercel and Supabase, with no virtual machine and no credit card.