FLASHBACK

Inspiration

Time passes, but memories remain. For Alzheimer’s patients, the struggle to recall core memories obscures the relationships and experiences that define them.
We researched reminiscence therapy and implemented its core theory into a digital application.
Our team developed a digital memory book integrated with AI-powered text-to-speech.
This app serves as a functional resource to help Alzheimer’s patients recall core memories via repetitive exposure, leveraging human-like audio to trigger recognition of loved ones.

Immerses Alzheimer's patients in their formative memories by allowing relatives to post photos with personalized, embedded text-to-speech.
Allows family members to communicate through a digital feed designed to mimic a traditional memory book with descriptive captions.
Features AI-powered audio on each image that precisely mirrors the relative’s actual cloned voice.
Combines visual and auditory stimuli to offer a more robust method for memory training and recollection than a phone call or physical book alone.

We created FLASHBACK using a lightweight, fast, and accessible full-stack approach:
Frontend: Built entirely with vanilla HTML, CSS, and JavaScript for rapid load times and complete DOM control without bulky frameworks.
Backend: Node.js and Express were used to deploy a secure server proxy that handles API calls and cross-origin requests.
AI Voice Integration: We integrated the ElevenLabs API to power the core auditory experience, dynamically cloning voices and generating high-fidelity text-to-speech linked to specific photos.
Database: Utilized Firebase Firestore to manage user accounts, store memory data, and enforce a robust role-based access system between patients and relatives.

API Security (CORS): Direct calls to the ElevenLabs API from the browser were blocked by CORS policies. We had to pivot and engineer a custom Node.js backend proxy to safely route requests and hide API keys.
Asynchronous audio management: High-quality AI voice generation takes a few seconds, leading to overlapping audio tracks when users spammed the "Play" button. We engineered a custom audio tracking system with generation IDs to lock the UI, prevent "ghost plays," and destroy abandoned fetches.
Database structuring: Seamlessly binding unique API keys and cloned Voice IDs to specific memories in Firebase without breaking access controls required complex data tunneling and state management.

Successfully built a seamless role-based architecture that allows family members to securely tunnel into a loved one's profile and narrate memories from anywhere.
Engineered a custom, highly accessible UI from scratch, featuring a high-contrast theme and dark mode to assist elderly patients with visual impairments.
Flawlessly integrated dynamic ElevenLabs voice cloning to instantly recognize and play the correct relative's voice for each specific memory.
Bridged complex AI backend tools with a beautifully simple, comforting, and empathetic frontend interface.

Backend networking: We gained deep hands-on experience in full-stack integration, secure API routing, and building custom proxies for sensitive credentials.
State management: We learned how to effectively manage vanilla JavaScript state and keep a dynamic UI perfectly synced with a live Firebase database without relying on frameworks like React.
Web Audio API: We learned how to safely stream, track, and terminate audio blobs in the browser to handle edge-case user inputs.
Accessible design: We learned how to balance powerful AI capabilities with strict, real-world accessibility standards and empathetic design.

IoT Integration: We plan to expand FLASHBACK beyond the screen by integrating it into physical devices like digital picture frames or simple push-button interfaces.
LLM AI Assistants: We are looking into integrating Large Language Models (like Claude) to analyze uploaded photos and automatically suggest warm, context-aware captions for relatives.
Expanded capabilities: We want to add multi-language support and allow for multi-voice conversational memories on a single photo.
Clinical testing: We plan to partner with reminiscence therapists to gather clinical feedback and refine the healing experience.