DIA26

DeepFender : Smart Email Guardian ! Detect, Explain, and Defend Against Phishing in Real-Time

Comment

💡 Inspiration Phishing is no longer just a corporate risk—it’s a daily threat for individuals, small businesses, and even students. With increasingly sophisticated social engineering and AI-generated scams, we wanted to build a defense system that feels like having a cybersecurity expert sitting in your inbox. Our goal: make anti-phishing proactive, intelligent, and user-friendly.

⚙️ What it does DIA26 (DeepFender) is a browser-based intelligent assistant that flags suspicious emails directly inside Gmail. It works by:

Continuously scanning your inbox for suspicious patterns.

Flagging high-risk emails with an inline visual warning.

Letting users ask “Why?” and receive natural explanations.

Connecting to a local AI agent server (FastAPI) that analyzes emails using multiple specialized agents powered by LLMs.

Offering a chatbot interface to help users understand risks in plain language.

🛠️ How we built it Gmail Content Script (Chrome Extension): Injects into Gmail, detects and highlights risky emails in real-time.

FastAPI Backend: Acts as a local intelligent agent hub. It receives email content and returns:

A risk score

A summary

A reason for flagging

Python Agents: Simulate an intelligent AI team:

InboxScanner Agent

PhishDetect Agent (LLM)

LinkVerifier Agent

TrustProfiler Agent

Decision Agent

Mistral AI LLM API: Handles the reasoning behind phishing detection and user chat explanations.

Local Gmail Listener (Optional): Retrieves new email content using the Gmail API for advanced scanning and automation.

🧱 Challenges we ran into Managing cross-context communication between the Gmail UI and the backend agent logic.

Designing a chatbot UX that feels native to Gmail but operates independently.

Making LLM responses feel reliable, safe, and contextual—especially when dealing with real emails.

Working with limited time, no production email datasets, and needing to simulate realistic phishing logic.

Ensuring everything stays local-first and privacy-aware.

🏆 Accomplishments that we're proud of Seamless Gmail overlay that doesn't disrupt the normal experience.

Modular AI agent architecture with room to scale to other platforms (Slack, SMS, etc.)

A chatbot that explains security in human terms, not tech jargon.

Custom FastAPI backend with fake but believable phishing detection simulation.

Completed a full "inbox to insight" loop with full-stack AI coordination.

🧠 What we learned How to integrate LLMs responsibly into user-facing tools.

That phishing detection isn't binary—it requires combining multiple subtle signals.

How much users appreciate explanations, not just warnings.

That security UX matters just as much as security logic.

🚀 What's next for DIA26 ✅ Add real-time phishing training mode to simulate safe/unsafe emails.

🧠 Upgrade LLM agents to handle multiple languages and tone detection.

🧪 Add actual sandbox testing for link verification via browser automation.

đź”— Expand to other platforms: Gmail mobile, Outlook, Discord.

🛡️ Create a Trust Score dashboard based on a user’s history of phishing exposure.

Built With

Share this project:

Updates