🧠 Remember Me

Smart glasses assistant for memory support — Dragonhack 2026

Built in under 24 hours at Dragonhack 2026 — from idea to working prototype.

⚠️ This is a hackathon prototype, not a production system. Expect rough edges.

---

📌 Overview

Remember Me is a prototype system designed to assist people with Alzheimer's disease and dementia by helping them recognise individuals and recall recent interactions.

When a person is detected through the camera, the system displays contextual information as an AR-style overlay:

• Name and relationship of the person
• Age and diagnosis context
• Summary of the last conversation
• Current time, date, and location

The goal is to reduce social stress and improve independence for people with memory impairment.

---

🏗️ Current Status

All core services are implemented and connected end-to-end:

• ✅ Frontend — React + Vite app with live face detection and overlay UI
• ✅ Backend — Express API orchestrating frontend ↔ ML service communication
• ✅ ML Service — Python FastAPI service generating FaceNet face embeddings
• ✅ Database — PostgreSQL Dockerized with schema and seed data

What is not yet complete: real embedding-to-database matching. The backend currently returns stub data. The ML service generates real embeddings — they just are not compared against the database yet. That is the next step.

---

🧱 Architecture

Browser (camera)
  │
  │  face crop (base64 JPEG, every 2s)
  ▼
Backend :3001  ──── face image (multipart) ──▶  ML Service :8000
  │                                                  │
  │  ◀──── 128-dim FaceNet embedding ────────────────┘
  │
  │  similarity check
  ▼
PostgreSQL :5432
  │
  │  person data (name, relationship, last interaction…)
  ▼
Backend → Frontend → overlay rendered on video feed

Frontend captures the camera stream, runs TinyFaceDetector in-browser to locate faces, crops them, and sends crops to the backend on a 2-second throttle.

Backend is the orchestrator. It receives crops from the frontend, forwards them to the ML service, will eventually query the database for the closest embedding match, and returns structured person data.

ML Service does the neural network work. It accepts a face image and returns a 128-dimensional vector (embedding) that numerically represents that face. Similar faces produce similar vectors — this is what makes matching possible.

Database stores person profiles, their face embeddings, and interaction history.

---

🌐 Browser Requirement

A Chromium-based browser is required. This means Chrome, Edge, Brave, Arc, or any other browser built on Chromium.

The face detection model (face-api.js) and the camera APIs used by the frontend rely on browser features that are either unavailable or significantly less performant in Firefox and Safari. Using a non-Chromium browser will likely result in the camera feed not working or face detection failing silently.

---

💻 Running locally (for development)

Running services individually is easier when you are actively developing, because you get faster hot-reloading and direct access to logs.

Prerequisites

• Node.js ≥ 18
• Python 3.11 (not 3.12+ — TensorFlow does not yet support those)
• Docker (for PostgreSQL only)

1. Database

docker compose up -d db

2. ML Service

cd ml-service
python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

3. Backend

cd backend
npm install
npm run dev

4. Frontend

cd frontend
npm install
npm run dev

The frontend is available at http://localhost:5173.

---

🗄️ Database

We use PostgreSQL 16, running in Docker. See data/README.md for details on the schema and how initialisation scripts work.

Useful commands

Connect to the database inside the container:

docker exec -it remember_me_db psql -U remember_me_user -d remember_me

---

📁 Project Structure

remember-me/
├── frontend/          # React + Vite + face-api.js
├── backend/           # Node.js Express API
├── ml-service/        # Python FastAPI + DeepFace + FaceNet
├── data/              # SQL initialisation scripts (schema + seed)
├── scripts/           # Utility scripts (model download, etc.)
└── docker-compose.yml # Full stack orchestration

---

👥 Team

Built at Dragonhack 2026 in under 24 hours.

🧠 Remember Me

Smart glasses assistant for memory support — Dragonhack 2026

Built in under 24 hours at Dragonhack 2026 — from idea to working prototype.

⚠️ This is a hackathon prototype, not a production system. Expect rough edges.

---

📌 Overview

Remember Me is a prototype system designed to assist people with Alzheimer's disease and dementia by helping them recognise individuals and recall recent interactions.

When a person is detected through the camera, the system displays contextual information as an AR-style overlay:

• Name and relationship of the person
• Age and diagnosis context
• Summary of the last conversation
• Current time, date, and location

The goal is to reduce social stress and improve independence for people with memory impairment.

---

🏗️ Current Status

All core services are implemented and connected end-to-end:

• ✅ Frontend — React + Vite app with live face detection and overlay UI
• ✅ Backend — Express API orchestrating frontend ↔ ML service communication
• ✅ ML Service — Python FastAPI service generating FaceNet face embeddings
• ✅ Database — PostgreSQL Dockerized with schema and seed data

What is not yet complete: real embedding-to-database matching. The backend currently returns stub data. The ML service generates real embeddings — they just are not compared against the database yet. That is the next step.

---

🧱 Architecture

Browser (camera)
  │
  │  face crop (base64 JPEG, every 2s)
  ▼
Backend :3001  ──── face image (multipart) ──▶  ML Service :8000
  │                                                  │
  │  ◀──── 128-dim FaceNet embedding ────────────────┘
  │
  │  similarity check
  ▼
PostgreSQL :5432
  │
  │  person data (name, relationship, last interaction…)
  ▼
Backend → Frontend → overlay rendered on video feed

Frontend captures the camera stream, runs TinyFaceDetector in-browser to locate faces, crops them, and sends crops to the backend on a 2-second throttle.

Backend is the orchestrator. It receives crops from the frontend, forwards them to the ML service, will eventually query the database for the closest embedding match, and returns structured person data.

ML Service does the neural network work. It accepts a face image and returns a 128-dimensional vector (embedding) that numerically represents that face. Similar faces produce similar vectors — this is what makes matching possible.

Database stores person profiles, their face embeddings, and interaction history.

---

🌐 Browser Requirement

A Chromium-based browser is required. This means Chrome, Edge, Brave, Arc, or any other browser built on Chromium.

The face detection model (face-api.js) and the camera APIs used by the frontend rely on browser features that are either unavailable or significantly less performant in Firefox and Safari. Using a non-Chromium browser will likely result in the camera feed not working or face detection failing silently.

---

💻 Running locally (for development)

Running services individually is easier when you are actively developing, because you get faster hot-reloading and direct access to logs.

Prerequisites

• Node.js ≥ 18
• Python 3.11 (not 3.12+ — TensorFlow does not yet support those)
• Docker (for PostgreSQL only)

1. Database

docker compose up -d db

2. ML Service

cd ml-service
python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

3. Backend

cd backend
npm install
npm run dev

4. Frontend

cd frontend
npm install
npm run dev

The frontend is available at http://localhost:5173.

---

🗄️ Database

We use PostgreSQL 16, running in Docker. See data/README.md for details on the schema and how initialisation scripts work.

Useful commands

Connect to the database inside the container:

docker exec -it remember_me_db psql -U remember_me_user -d remember_me

---

📁 Project Structure

remember-me/
├── frontend/          # React + Vite + face-api.js
├── backend/           # Node.js Express API
├── ml-service/        # Python FastAPI + DeepFace + FaceNet
├── data/              # SQL initialisation scripts (schema + seed)
├── scripts/           # Utility scripts (model download, etc.)
└── docker-compose.yml # Full stack orchestration

---

👥 Team

Built at Dragonhack 2026 in under 24 hours.

🧠 Remember Me

Smart glasses assistant for memory support — Dragonhack 2026

Built in under 24 hours at Dragonhack 2026 — from idea to working prototype.

⚠️ This is a hackathon prototype, not a production system. Expect rough edges.

---

📌 Overview

Remember Me is a prototype system designed to assist people with Alzheimer's disease and dementia by helping them recognise individuals and recall recent interactions.

When a person is detected through the camera, the system displays contextual information as an AR-style overlay:

• Name and relationship of the person
• Age and diagnosis context
• Summary of the last conversation
• Current time, date, and location

The goal is to reduce social stress and improve independence for people with memory impairment.

---

🏗️ Current Status

All core services are implemented and connected end-to-end:

• ✅ Frontend — React + Vite app with live face detection and overlay UI
• ✅ Backend — Express API orchestrating frontend ↔ ML service communication
• ✅ ML Service — Python FastAPI service generating FaceNet face embeddings
• ✅ Database — PostgreSQL Dockerized with schema and seed data

What is not yet complete: real embedding-to-database matching. The backend currently returns stub data. The ML service generates real embeddings — they just are not compared against the database yet. That is the next step.

---

🧱 Architecture

Browser (camera)
  │
  │  face crop (base64 JPEG, every 2s)
  ▼
Backend :3001  ──── face image (multipart) ──▶  ML Service :8000
  │                                                  │
  │  ◀──── 128-dim FaceNet embedding ────────────────┘
  │
  │  similarity check
  ▼
PostgreSQL :5432
  │
  │  person data (name, relationship, last interaction…)
  ▼
Backend → Frontend → overlay rendered on video feed

Frontend captures the camera stream, runs TinyFaceDetector in-browser to locate faces, crops them, and sends crops to the backend on a 2-second throttle.

Backend is the orchestrator. It receives crops from the frontend, forwards them to the ML service, will eventually query the database for the closest embedding match, and returns structured person data.

ML Service does the neural network work. It accepts a face image and returns a 128-dimensional vector (embedding) that numerically represents that face. Similar faces produce similar vectors — this is what makes matching possible.

Database stores person profiles, their face embeddings, and interaction history.

---

🌐 Browser Requirement

A Chromium-based browser is required. This means Chrome, Edge, Brave, Arc, or any other browser built on Chromium.

The face detection model (face-api.js) and the camera APIs used by the frontend rely on browser features that are either unavailable or significantly less performant in Firefox and Safari. Using a non-Chromium browser will likely result in the camera feed not working or face detection failing silently.

---

💻 Running locally (for development)

Running services individually is easier when you are actively developing, because you get faster hot-reloading and direct access to logs.

Prerequisites

• Node.js ≥ 18
• Python 3.11 (not 3.12+ — TensorFlow does not yet support those)
• Docker (for PostgreSQL only)

1. Database

docker compose up -d db

2. ML Service

cd ml-service
python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

3. Backend

cd backend
npm install
npm run dev

4. Frontend

cd frontend
npm install
npm run dev

The frontend is available at http://localhost:5173.

---

🗄️ Database

We use PostgreSQL 16, running in Docker. See data/README.md for details on the schema and how initialisation scripts work.

Useful commands

Connect to the database inside the container:

docker exec -it remember_me_db psql -U remember_me_user -d remember_me

---

📁 Project Structure

remember-me/
├── frontend/          # React + Vite + face-api.js
├── backend/           # Node.js Express API
├── ml-service/        # Python FastAPI + DeepFace + FaceNet
├── data/              # SQL initialisation scripts (schema + seed)
├── scripts/           # Utility scripts (model download, etc.)
└── docker-compose.yml # Full stack orchestration

---

👥 Team

Built at Dragonhack 2026 in under 24 hours.

• Zhivko Stoimchev
• Pavel Jolakoski
• Hristijan Chupetreski
• Nikola Murgovski