$ cd ~/projects/deafeat
Deafeat
A hardware device and a trained machine learning model that translate sign language into text and speech, and back, for two-way communication between deaf and hearing users.
The Deafeat AI hardware during a progress-update review.
Overview
Deafeat AI is a system that detects hand gestures and translates them into letters and words, which are then combined into sentences and, finally, speech. It also works in reverse, translating spoken language into text, for two-way communication between deaf and hearing people.
My Role
I founded the project and lead a six-person team. I work on the machine learning model, the training pipeline, and the hardware integration alongside one other engineer, in addition to the technical direction and team leadership.
How It Works
- Camera integration with frame-rate optimization, and library experimentation to handle inconsistent lighting
- A static sign-recognition model integrated with a dynamic gesture model in a unified inference pipeline
- Temporal logic for tracking gestures over time, plus geometric preprocessing to improve detection accuracy
- A lightweight custom-training menu with automated data augmentation, to work around a limited training dataset
- Deployed on a Raspberry Pi, within its limited processing power
Testing
Deafeat has been tested at the Lebanese School for the Blind and the Deaf (LSBD) in Baabda, Lebanon in May 2025.
Status & Recognition
The project is copyright registered with the Lebanese Ministry of Economy, and a USPTO patent application is in preparation. Deafeat has been recognized by NuVuX, a US-based innovation studio affiliated with MIT, and its Instagram community has grown to over 3,000 followers, most of whom are deaf. The team also received a Social Entrepreneurship award for Deafeat at International College.
What's Next
Planned directions include a standalone mobile app, hardware redesigns, and support for additional sign languages.