ReminAIs

Inspiration

Dementia is a thief that steals not just memories, but connection. For the millions of people living with Alzheimer's and dementia, the anxiety of not recognizing a loved one, a daughter, a lifelong friend, a grandchild, can be paralyzing. This anxiety often leads to social withdrawal and profound isolation.

We wanted to build a bridge. ReminAiS (Reminisce + AI) acts as a "digital hippocampus," whispering context into the user's ear the moment a visitor arrives, turning a moment of confusion into a moment of connection.

What it does

ReminAiS is an intelligent companion app that runs on a tablet or a wearable device.

Instant Recognition: When a visitor enters the room, the app uses facial recognition to identify them instantly. Contextual Whisper: It doesn't just say a name. Using Google Gemini, it generates a warm, personalized greeting based on the visitor's bio and past conversations (e.g., "This is Pierre, your grandson. Last time, you talked about his soccer game"). Human-Like Voice: The greeting is read aloud using ElevenLabs' ultra-realistic text-to-speech, providing a comforting, familiar presence. Memory Timeline: It listens to the conversation (Speech-to-Text), summarizes it, and saves it to a timeline. This helps the patient recall what they talked about previously, reinforcing memory pathways. Smart Dashboard: A mobile-first interface for patients to view their connections and look back on happy memories.

How we built it

We built ReminAiS with a modern, privacy-focused stack:

Frontend: Built with React and Vite for speed. We used Tailwind CSS to create a highly accessible, high-contrast UI suitable for elderly users. Computer Vision: We implemented face-api.js to perform face detection and recognition entirely in the browser. This ensures sensitive biometric data doesn't necessarily need to leave the device. The Brain (AI): We leveraged Google Gemini Pro to process the context. We feed it the identified person's bio and the history of previous chats to generate natural, reassuring prompts. Voice: Input: react-speech-recognition for transcribing live conversations. Output: ElevenLabs API for low-latency, emotionally resonant speech synthesis. Data Persistence: Local storage (prototyping) to keep connection history and bios.

Challenges we ran into

Real-time Performance: Running face detection alongside a webcam feed in the browser can be heavy. We had to optimize the recognition interval to balance accuracy with battery life/performance. Context Window: Feeding the entire conversation history to the LLM can get expensive and slow. We implemented a summarization step where Gemini condenses a conversation into a short memory before saving it to the database. Hardware Access: Managing permissions for Camera and Microphone simultaneously while switching between "Admin" and "Patient" modes required careful state management in React.

Accomplishments that we're proud of

Edge Intelligence: Face detection is loaded locally to make sure the core recognition functions work even when the internet connectivity goes down.
Accessible UI: We worked on a professional design that would fill the requirements of senior users in terms of accessibility without removing the modern touch.
Prompt Engineering: In the instructions for Gemini, we emphasize empathy, so that the AI says, "Look, it's your friend." rather than just stating a name.

What we learned

We enhanced our knowledge of "Edge AI" and its value in "Privacy by Design" when dealing with biometric information. Finally, we were introduced to how one can "fine-tune the emotional temperature" of an "LLM" to make sure one feels comforted, not merely informed.

What's next for ReminAIs

Sentiment Analysis: Tracking the patient's mood over time based on facial expressions during interactions.
Wearable Integration: Moving from a tablet to smart glasses (like Ray-Ban Meta or similar) for a seamless heads-up display.
Cloud Sync for Families: An app for family members to upload their own photos and "life updates" remotely, which get fed into the patient's context for their next visit.