Inspiration
Alzheimer's disease affects over 55 million people worldwide, and one of its most heartbreaking symptoms is the loss of facial recognition — the moment a patient looks at their own child and sees a stranger. We wanted to do something about that. The idea came from thinking about how powerful modern AI has become at recognizing faces, and asking ourselves: what if we put that directly in front of someone's eyes? Meta smart glasses gave us the perfect canvas.
What it does
Nazr is a wearable memory companion built into Meta smart glasses. When an Alzheimer's patient looks at a familiar face, the glasses instantly recognize who that person is and discreetly deliver three key pieces of context through audio: their name, their relationship to the patient, and a summary of their last conversation together. No screens, no buttons, no caregiver needed — just a quiet, dignified reminder delivered right when it's needed most. It also recognizes if the user has drank water, or eaten after a certain amount of time, while simultaneously updating a caregiver dashboard.
How we built it
We used Meta smart glasses as the hardware platform, leveraging the built-in camera to capture the wearer's field of view in real time. That footage is sent to the Gemini API, which handles facial recognition and generates a natural language response with the relevant context. Recognized faces are matched against a stored profile database containing each person's name, relation, and conversation history. The response is then delivered back to the wearer through the glasses' built-in speakers.
Challenges we ran into
Latency — getting the recognition pipeline fast enough to feel natural was one of our biggest hurdles. A few seconds of delay can completely break the moment. Real-world accuracy — facial recognition in uncontrolled lighting conditions, angles, and movement is significantly harder than it looks in demos.
Accomplishments that we're proud of
Built a fully working end-to-end demo — from camera feed to recognized face to audio output — within the hackathon timeframe Successfully integrated the Gemini API for real-time vision and language tasks Built something with genuine real-world impact for one of the most underserved populations in tech
What we learned
How to work with the Gemini API for real-time vision tasks under time pressure That the hardest part of assistive tech isn't the AI — it's the UX: making it feel natural, fast, and trustworthy
What's next for Nazr
Expanding the profile system — letting caregivers and family members easily add and update people through a companion app Reducing latency further through optimized on-device processing
Log in or sign up for Devpost to join the conversation.