Rememberly: Helping dementia patients reconnect with their cherished moments.

Source: Dev.to

Overview

Rememberly is an AI‑powered memory companion designed for families caring for loved ones with dementia and memory loss. It transforms scattered photos, videos, voice notes, and written stories into an interactive memory library with two distinct interfaces:

• For families & caregivers – a simple dashboard to upload, tag, and organize digital memories in a shared library.
• For patients – a gentle, conversational interface that lets them ask questions, browse memories, and receive AI‑generated “Cinematic Shows” that surface relevant moments from their life.

Both parts are powered by Mistral (for reasoning and content generation) and Vectara (for secure retrieval‑augmented generation).

How It Works

Caregiver Workflow

1. Upload media (images, videos, text, voice notes).
2. The system extracts tags and generates descriptions using Mistral.
3. Tagged documents are ingested into a patient‑specific knowledge box in Vectara.
4. Media and metadata are stored securely in Convex.

Patient Workflow

1. Interact via natural language (text or speech).
2. The AI agent retrieves relevant memories from the patient’s isolated corpus.
3. Responses may be plain text or immersive cinematic presentations that combine images, audio, and generated narration.

Technical Architecture

Sequence Diagram

sequenceDiagram
    participant Caregiver
    participant Patient
    participant Convex
    participant Mistral
    participant Vectara
    participant Deepgram

    Note over Caregiver,Vectara: Media Upload Flow
    Caregiver->>Convex: Upload media (images/videos/text)
    Convex->>Mistral: Extract tags & generate description
    Mistral-->>Convex: Return tags & description
    Convex->>Vectara: Ingest document to patient corpus
    Convex->>Convex: Store media with tags

    Note over Patient,Deepgram: Retrieval & Interaction Flow
    Patient->>Convex: Query memory (text or voice)
    Convex->>Deepgram: Speech‑to‑text (if voice)
    Convex->>Vectara: Retrieve relevant documents
    Vectara-->>Convex: Return matched memories
    Convex->>Mistral: Generate response / cinematic show
    Mistral-->>Patient: Return answer (text or multimedia)

Core Components

All patient data is kept private and isolated; each patient has a dedicated knowledge box in Vectara.

Testing Instructions

Option 1 – New Patient Onboarding

1. Create a new patient account.
2. Upload pictures, videos, or text to build the personal memory library.
3. Interact with the patient interface to test the full workflow (upload → retrieval → cinematic show).

Option 2 – Pre‑configured Test Accounts

• Email: james5@gmail.com
• Password: qahket-pumde6-Rosdub

Use these credentials to explore both caregiver and patient experiences without setting up a new account.

Live Project

Live Project Link

Source Code

View on GitHub