Inspiration
We were inspired by the growing problem of digital fraud. Fake accounts, stolen identities, and duplicate registrations are common challenges faced by banks, fintech companies, and even government services. We wanted to create a solution that not only prevents fraud but also makes identity verification simple and trustworthy for users.
What it does
Our project uses Agentic AI to combine face deduplication with liveness detection. It checks if a user’s face is already in the database (to prevent duplicate accounts) and confirms that the selfie is from a live person, not a photo or video. The system adapts over time, learning from new cases to improve accuracy and stay ahead of fraud tactics.
How we built it
We built the solution with a mobile-first approach. A user takes a selfie, which is processed by a deep learning model to generate a unique facial embedding. Another AI module performs a liveness test by asking the user to blink, smile, or turn their head. These embeddings are then compared with the database in real time using secure cloud-based GPU services. We used Python, TensorFlow, OpenCV, and FaceNet for model development, with APIs to connect the backend AI to the app.
Challenges we ran into
a)Balancing security with a smooth user experience. b)Handling large datasets efficiently while keeping latency low. c)Training liveness models to detect advanced spoofing methods like videos and masks. d)Ensuring compliance with data privacy and security standards.
Accomplishments that we're proud of
a)Designed a working pipeline that integrates both deduplication and liveness in real time. b)Achieved accurate detection with minimal false positives. c)Created a scalable solution that can handle millions of users. d)Built a user-friendly flow that requires only a simple selfie.
What we learned
We learned how powerful agentic AI can be when applied to fraud detection. We also understood the importance of designing systems that are both technically strong and user-friendly. Most importantly, we realized that trust is at the heart of identity verification.
What's next for Agentic AI for Face Deduplication and Liveness Verification
We plan to: a)Expand liveness detection with multimodal signals (voice, movement, multispectral sensors). b)Optimize the pipeline for edge devices so verification can work offline. c)Test the system with real-world datasets to improve accuracy. d)Explore integration with industries like fintech, healthcare, and e-commerce for broader adoption.
Log in or sign up for Devpost to join the conversation.