WOKE | Devpost

home page!
getting a call
calling!

Inspiration

In the era where AI is increasingly prevalent, it's not uncommon to encounter its misuse. From fraudulent schemes to malicious impersonation, the risks posed by AI exploits are real and impactful. For instance, in Hong Kong, a deepfake video duped a firm out of $25 million dollars, highlighting the potential dangers. Recognizing the vulnerability of older generations to such deceptive technologies, our inspiration stems from a genuine concern for the well-being of individuals. Our aim is to contribute towards advancing the 16th Sustainable Development Goal (SDG) of Peace, Justice, and Strong Institutions by combating the threat posed by deepfakes.

What it does

WOKE helps you stay woke against the perils of deepfake manipulation. It operates as a secure video-calling service equipped with AI capabilities to detect instances of deepfake impersonation. Seamlessly integrated into the user experience, WOKE utilizes Gemini's AI model to analyze incoming video streams in real-time. Upon detection of any signs of deepfake manipulation, WOKE promptly alerts the user, providing vital warnings and safeguards. Leveraging our model, various frames of the call recipient's face are scrutinized by Gemini, utilizing Google's powerful infrastructure to ascertain conversational authenticity.

How we built it

WOKE uses a flask backend, connected to a React frontend while utilizing tools like Google Gemini AI and Google Cloud Storage to temporarily store, process, and determine whether the other caller is a deepfake.

Challenges we ran into

24 hours is an incredibly short amount of time. We wanted to implement more features like a transcription model post-call that acts as another form of verification as to whether the other caller is a deepfake. We also faced several issues when trying to implement Google's APIs and deploying our project to use the call feature on 2 separate devices.

Accomplishments that we're proud of

The amount of work we were able to get done in just 24 hours! In just 24 hours, we were able to prototype a working call feature, detect deepfakes, and use 2 Google services: Gemini AI and Google Cloud Storage.

What we learned

We learned a lot about generative AI and the intricacies of AI integration and real-time data processing. Navigating the complexities of deploying third-party APIs underscored the significance of thorough documentation and effective troubleshooting strategies. Throughout this project, we also looked at various research studies and papers to help us better prompt Gemini and to learn more about deepfakes in general.

What's next for WOKE

Transcription model to also detect deepfake audio. Smoother UX and additional personalization features. Further software and AI development. Turn WOKE into a mobile app to also detect if phone calls are deepfakes.