REMIND

• 

  mhmmmm

Inspiration

Last spring, a member of our team watched their grandfather get diagnosed with dementia, the same news fifty-seven million families around the world receive every year, with a new diagnosis every three seconds. He built his own house with his hands and taught his grandson football in the garden, and there is no cure for what's happening to his memory. What struck us wasn't just the disease, it was that medicine has spent forty years and billions of dollars fighting the biology and almost nothing fighting the other loss, the one where a person stops recognizing the people who love them. We realized dementia takes someone twice, once in the brain and once in the world, and while the first loss might not be solvable yet, we became convinced the second one is. That's the idea we built REMIND around.

What it does

REMIND is a pair of prescription AI glasses for people living with dementia. A doctor sets them up with the family, and they run on the wearer's own daily rhythm, waking when they wake and resting when they rest. Through the day, they recognize approved faces and whisper who someone is before the panic can set in, they give gentle reminders for things like medication without ever nagging, and they pass along messages from family and doctors at calm moments. Every evening, they hand the whole day back as a warm, simple recap, so nothing lovely gets lost even if it's forgotten. If the wearer gets disoriented outside, the glasses guide them home. If a hard moment comes, they play a familiar voice or song to bring calm. And on the clinical side, doctors get a real picture of how the patient is doing between appointments instead of a single snapshot every six months, while families control exactly what's shared, so privacy stays a choice rather than a setting nobody checks.

How we built it

We built REMIND's demo pipeline around the Groq API for fast inference, using it to power the recognition logic and the natural language generation behind the evening recaps and reminder messages, since Groq's speed made it realistic to simulate real-time responses instead of noticeable lag every time the glasses "recognized" someone. For the face scanning piece, we used a face detection and embedding model to identify approved faces from camera input and match them against a small consented gallery, which is what powers the moment where the glasses recognize a family member and whisper who they are. We stored and retrieved memories, recap data, and matched faces using vector embeddings in MongoDB Atlas, so a face or a moment could be looked up by similarity rather than an exact match, which is closer to how the real system would need to work. On the front end we built a fully interactive web prototype simulating the wearer's lens view, the clinician dashboard, the family review queue, and the scheduling system, all connected to the same underlying data so the demo tells one consistent story instead of a bunch of disconnected screens. We used the Groq API again for the speech-pattern analysis behind Cognitive Weather, generating simulated daily trend data that shows how the system would flag cognitive changes over time. Everything was designed to be swapped for real on-device hardware later, but for the hackathon we prioritized proving the full loop end-to-end, camera input in, recognition and redaction happening, memory stored, and the right message reaching the right person, over any single piece being production-ready. (Prototype)

Challenges we ran into

Honestly, the hardest problem for us wasn't technical at all, it was figuring out how to get the emotional side right, and that meant asking one question over and over again for every single feature we designed, does this protect someone's dignity or does it quietly take it away, and that question ended up killing a lot of ideas that would have been way easier to build but felt invasive, patronizing, or clinical in a way that we knew a person actually living with dementia would hate. Privacy turned out to be its own whole challenge too, because dementia care often means family members need to see private moments to actually help, but we really didn't want to default to constant surveillance, so instead we built a system where the wearer's family reviews things themselves and chooses what gets shared, rather than just logging everything automatically behind their back. On top of that, we kept running into the challenge of making the interface feel like it wasn't even there, since the whole point of REMIND is that someone who can no longer learn new technology still needs to be able to use it without learning anything at all.

Accomplishments that we're proud of

We're proud that REMIND never feels like a health app pretending to have a personality, it feels like a person got their dignity back. We built a fully interactive prototype that lets anyone step into the wearer's exact point of view, see a real recognition moment happen, watch a day become a recap, and feel the privacy choice as it's made rather than reading about it in a spec sheet. We designed and justified support for twenty distinct forms of dementia rather than treating the disease as one thing, and we built a clinician view that actually gives doctors continuous signal instead of a guess every six months. Most of all, we're proud that every decision traces back to one real person, because that kept us honest about what actually helps versus what only looks impressive in a deck.

What we learned

We learned that the hardest part of building assistive technology isn't the technology, it's resisting the urge to build for the caregiver instead of the person. It's tempting to design dashboards and control panels because those are satisfying to build, but the actual user here can't operate anything, so every good decision we made came from asking what a person experiences in the moment rather than what a feature looks like on a slide. We also learned how much dementia varies, and that a system built only around Alzheimer's would fail most of the people who need it, which pushed us to think in terms of adaptable care profiles instead of one fixed product.

What's next for REMIND

We want to keep working on this after the hackathon. Right now REMIND is a prototype, and our goal is to turn it into something real, with actual hardware, real families, and real doctors using it, not just a demo. We also want to make sure it's accessible everywhere, since dementia affects families regardless of income or location, and this kind of support shouldn't be limited to people who can afford expensive medical technology. Privacy is another area we want to keep improving, because families need to feel like they're always in control of what's shared and what stays private, and we think there's still a lot of room to get that right.

Built With

• css
• groq
• html
• javascript