💡 About the Project

Inspiration

Alzheimer’s is a disease that slowly takes away one of life’s most precious gifts: our memories.
I was inspired to create Memoria after seeing how families struggle to support loved ones through memory loss.
Instead of another generic health app, I wanted to design a gentle, local-first companion — one that respects privacy, offers calm encouragement, and turns everyday moments into opportunities for memory practice.


What I Learned

  • 🧠 How to integrate a large local LLM (GPT-OSS:20B) into a memory care workflow
  • 🗃️ The importance of local storage (SQLite) for privacy-first healthcare apps
  • 🎨 Designing an interface that is calm, simple, and accessible for elderly users
  • 🔐 Why local-first AI matters: no internet dependency, no risk of sensitive data leaving the device

How I Built It

  1. Frontend (React + Tailwind)

    • A lightweight, accessible UI designed for simplicity.
    • Includes Zen Stones visualization for progress — a metaphor for stability and growth.
  2. Backend (Node.js + Express + SQLite)

    • Local API endpoints for logging, testing, and scoring memories.
    • Uses memories.db to securely store all personal data on-device.
  3. AI Engine (GPT-OSS:20B via Ollama)

    • Generates personalized quizzes from user-logged memories.
    • Evaluates answers in natural language, not just exact matches.
    • Provides AI Feedback: encouraging notes, tips, and progress insights.
  4. Alerts & Safety Layer

    • If average test performance drops below a safe threshold (default:
      [ \text{Alert Trigger} = \text{Average Score} < 50\% ]
      ), Memoria gently prompts caregiver notification.

Challenges I Faced

  • ⚖️ Model Optimization — Running GPT-OSS:20B locally required tuning performance while keeping memory usage safe.
  • 🖥️ Local-only Architecture — Ensuring zero cloud calls while still delivering AI-powered personalization.
  • 🎭 UX for Vulnerable Users — Designing an app that is calming, intuitive, and respectful for elderly users.
  • 💬 Balancing AI Feedback — Making responses empathetic and supportive, not clinical or robotic.

✨ In building Memoria, I learned that technology alone cannot solve Alzheimer’s — but it can provide hope, dignity, and connection. This project is a small step toward that vision.

Built With

Share this project:

Updates