GELATO: AI-Powered Cyber Vigilance with Italian Flair

Scams melt away like gelato under the sun ☀️ • Flavor Alerts - Risk levels decoded through gelato hues: 🍃 Pistachio Green (Security) 🌕 Mango Yellow (Caution) 🌶️ Fiery Pepper Red (Alert)

Comment

Inspiration

In 2015, while enjoying coconut gelato at Vancouver's White Rock Pier during summer camp, I clicked a phishing link disguised as "Help vote for my sister" from a hacked classmate’s account. A lock of mine social account and a frantic call from my mom moments sparked an idea: "Cybersecurity should be as intuitive as craving gelato on a hot day"—simple, delightful, and protective.

What It Does

GELATO (Guarding Everyone's Links Against Threatening Operations) is a Chrome extension that empowers users to surf safely with real-time scam detection:

  • Instant Text Analysis: Spots scams as you browse or select text.
  • Gelato Scoop: On-demand analysis with a playful gelato-themed UI.
  • Risk Flavors: Visual feedback with Pistachio (low), Mango (medium), and Pepper (high) risk levels.
  • Pattern Matching: Flags common scam tactics and suspicious phrases.
  • Case Comparison: Cross-references text with known scam cases.
  • Actionable Reports: Delivers detailed insights and safety tips.

For users, it’s like having a cyber-savvy friend watching over every click.

How We Built It

🧑💻 Crafting the Gelato Layers

  • ** Prototyping ：Cursor + OpenAI -Accelerated component iteration
  • **Data Engine ：NumPy + Scikit-learn-Synthetic data augmentation
  • **AI Core ：DeepSeek-RAG + FAISS-Custom prompt chaining for accuracy
  • **UI/UX ：Gelato shop blueprint- Risk-level flavor visualizations

Challenges We Faced

🔥 The Melting Point

  • API Freeze: Hit GPT-4 rate limits in final tests → pivoted to a paid API subscription.
  • Extension Overload: Full-page scans spiked CPU usage → switched to "Gelato Scoop" on-demand model.
    • Single RAG struggled with fresh scam patterns due to outdated datasets.
    • Internet lacks rich, balanced scam vs. non-scam data.
    • Adapting to diverse website layouts proved tricky.

Accomplishments We’re Proud Of

  • Robust Database: Built 1,200+ scam pattern entries for training.
  • Delightful UI: Gelato-inspired design with intuitive risk metaphors.
  • Real-World Impact: Detected 86% of phishing attempts in Gmail tests.
  • Platform Precision: Hit 92% accuracy on platforms like Gmail.

What We Learned

Technical Hurdles:

  • RAG’s memory limits in browser environments.
  • Chrome API integration quirks.

Dev Practices:

  • Git Lifeline: Hourly commits with auto-backups saved us.
  • Cost Hacks: 2AM-5AM API discounts cut costs by 50%.
  • Grit: Fixed 147+ bugs in 36-hour coding sprints.

What’s Next for GELATO

  • Data Growth: Expand database to 5,000+ entries.
  • Broader Reach: Boost site compatibility to 90% (from 65%).
  • Speed Surge: Slash latency from 1.2s to <0.5s.
  • Multimedia Scan: Prototype image-based link risk analysis.

Built With

Share this project:

Updates