GProtect

Inspiration

The inspiration behind GProtect came from the rising concerns around phishing emails, which have become one of the most common threats in cybersecurity. With more people relying on email for personal and professional communication, detecting and protecting against phishing scams is critical. We wanted to leverage AI to make email security more accessible and efficient for everyday users, helping them identify potentially harmful emails before they can cause harm.

What it does

GProtect uses advanced AI to detect phishing emails in real-time, providing users with an extra layer of security. By analyzing various aspects of incoming emails, such as the content, subject line, and sender details, the app can determine whether an email is legitimate or a phishing attempt. The app uses AI models fine-tuned specifically for this task, giving users confidence that their email inbox is safer from scams.

How we built it

We built GProtect using a combination of the Gmail API for email access, Gemini 1.5’s flash and fine-tuned models for phishing detection, and Kaggle’s phishing email dataset for training the AI. The app works by analyzing email data and using machine learning algorithms to classify emails as either phishing or safe. The fine-tuning process involved training the model on labeled phishing and legitimate email data, ensuring that the app could reliably identify phishing attempts with high accuracy.

Challenges we ran into

One of the main challenges we faced was ensuring the accuracy and reliability of the AI model. While there are many factors to consider, such as email structure, sender behavior, and content, fine-tuning the Gemini model with the Kaggle dataset helped us create a robust solution. Another challenge was integrating the Gmail API efficiently, ensuring seamless communication with users' inboxes. Additionally, keeping the model up-to-date with emerging phishing tactics posed a continuous hurdle.

Update made during this hackathon

Improved finetuning with more data.

Accomplishments that we're proud of

We are particularly proud of the accuracy of the phishing detection model. By fine-tuning the Gemini 1.5 model using a well-curated dataset, we’ve achieved strong performance in identifying phishing emails. We’re also proud of the user-friendly interface of GProtect, which makes it easy for anyone to use the app and receive real-time protection without needing deep technical knowledge.

What we learned

We learned a great deal about working with large AI models like Gemini 1.5 and fine-tuning them for specialized tasks. We also gained valuable experience in integrating APIs (like the Gmail API) into apps for real-world usage, as well as dealing with challenges around dataset quality, training techniques, and the evolving nature of phishing tactics. This project deepened our understanding of both machine learning and cybersecurity.

What's next for GProtect

Looking ahead, we plan to expand GProtect's capabilities by improving the model’s performance with more diverse datasets, including new phishing tactics. We also aim to introduce additional features, such as email reporting and customizable user settings, to enhance user experience. We are considering integrating GProtect with other email platforms beyond Gmail and exploring mobile app versions to protect users across all their devices.

Built With

• flask
• gemini-api
• gmail-api
• google-gmail-oauth
• nextjs
• python