A Datathon Tale: From Online to On‑Site
The preliminary round of the MCMC Datathon was held entirely online—our very first datathon (or hackathon‑style) adventure. We had zero expectations of making the finals, but to our delight, we did! And we manage to get a consolation prize!
The final showdown took place at the Amari Hotel Kuala Lumpur. Between technical challenges, we feasted on incredible hotel food and enjoyed the comfiest beds. I still daydream about those meals!
Project Overview
KidPhish is an innovative cybersecurity solution designed to protect children from accidentally accessing malicious links while using their parents’ devices. Using advanced deep learning techniques, the system can detect and block potentially harmful URLs in real-time.
Problem Statement
Children increasingly access online content through their parents’ devices, often without adequate supervision. This creates specific security vulnerabilities:
- Children may not recognize phishing attempts or malicious links
- Standard anti-phishing tools aren’t designed with child users in mind
- When children click on malicious links, parents’ sensitive information could be compromised
- Reporting mechanisms for harmful content are often too complex for young users
Solution
We developed a deep learning model specifically designed to detect malicious links that might target or be accidentally accessed by children. The solution includes:
- Real-time URL scanning and threat detection
- Child-friendly warning interfaces
- Simplified reporting workflow to key agencies
- Dashboard for parents to monitor and review blocked threats
Technologies Used
- TensorFlow for deep learning model development
- Machine Learning techniques for URL analysis
- Flask API for integration capabilities
- Containerized deployment for scalability
Integration Points
Our proposed solution included streamlined reporting workflows to key agencies:
- MCMC (Malaysian Communications and Multimedia Commission)
- National Scam Response Centre
- Cybersecurity Malaysia
Results
- Developed a working prototype during the 48-hour datathon
- Achieved 94% accuracy in detecting malicious URLs in our test dataset
- Received consolation prize recognition from judges
- Created a simplified technical implementation plan for potential adoption
Gallery
