Inspiration

Phishing remains one of the most common entry points for cyberattacks, often fooling even tech-literate users. As cybersecurity students passionate about AI, we wanted to build a tool that not only detects phishing messages but also educates users on why a message might be suspicious — empowering people through AI.

What it does

Our system analyzes any email or message and determines whether it’s “Phishing” or “Safe.” It uses an AI model to explain the reasoning behind its judgment, increasing user awareness. The interface is simple — paste your message, click analyze, and get an instant verdict with an explanation.

How we built it

Frontend: Streamlit web app for user interaction

Backend: Python-based API using FastAPI

AI Component: Prompt-based classification using a large language model (LLM)

Bonus Features: Keyword highlighter for common phishing triggers (e.g., “urgent”, “verify”, “account”)

Challenges we ran into

Crafting prompts that consistently return accurate classifications from the LLM

Avoiding false positives for legitimate emails with urgency (e.g., job offers, password resets)

Managing API rate limits and fallback mechanisms when the AI service is down

Balancing accuracy, speed, and simplicity for real-time use

Accomplishments that we're proud of

Built a fully functional phishing detection system in limited time

Designed a prompt that performs consistently without fine-tuning

Created a solution that goes beyond detection — it teaches

Made the UI friendly and deployable locally

What we learned

The power of LLMs in real-world security applications

The importance of prompt engineering

How to explain technical AI decisions to non-technical users

That cybersecurity + AI is a perfect blend for innovation

What's next for AI-Powered Phishing Detection System

Integrate with Gmail and Outlook for real-time scanning

Add support for attachments and links

Expand to multilingual phishing detection

Build a dashboard for tracking phishing trends over time

Open-source it for educational and enterprise use

Share this project:

Updates