PhiShield

PhiShield: Your Guardian Against The Major Email Threat

Comment

Inspiration

Everything starts a few months ago...

Farley and Gustavo, with a background in cyber security, had long contemplated building something in this field. After getting in touch with Rafael and discovering a shared vision for addressing critical cyber security issues, they joined forces to tackle what some already considered one of the world's greatest cyber crimes.

What it does

PhiShield is a software developed with Gmail users in mind. It's designed to be easy to set up and use. The software enables users to analyze their emails to detect potential attacks and report suspicious emails to help protect the community.

How we built it

Harnessing the power of Gemini, Generative AI, and our development skills, we take the user's email and segment it into multiple pieces of data. This data is then processed by our AI model, which receives the email header, content, links, attachments, sender etc. after receiving this data the LLMs evaluate the risk of the email.

Challenges we ran into

One significant challenge we faced was disabling the safety settings or guardrails from the AI model. This required careful consideration and testing to ensure the model's effectiveness.

Risk in development: we wanted to leverage HTML code information and images from sites, however the proper setup for development would require some time. Since the local/cloud development would pose some risks we decided to do not do this improvement right now, However we know this is possible and it would SOTA (state of the art) results for our solution

Accomplishments that we're proud of

We're particularly proud of the results produced by our AI model, especially its interpretability score, which helps us identify social engineering issues effectively. This capability allows us to address complex cyber security, for example, social engineering threats more effectively.

What we learned

Throughout the project, we learned valuable lessons about cyber security, AI model development, and collaboration. These lessons have helped us refine our approach and develop a more robust solution.

What's next for PhiShield

In the future, we plan to improve our AI models using Google Cloud, integrating our software into Google Workspace, improve the predictive performance by using screenshot from pages to process image information for multimodal models, usage of HTML code as feature and expand our range of features... and, who knows, maybe get bought by Google.

Built With

Share this project:

Updates