Dermafind is a full-stack acne analysis and skincare recommendation platform that combines computer vision with large language models to provide personalized skincare guidance based on facial image analysis.

The system uses a custom YOLOv26 model to detect acne types from uploaded images and an LLM-driven recommendation engine aligned with the IGA acne severity grading scale.

---

Features

• Image-based acne detection and classification
• Detection of six acne types:
  • Pustule
  • Dark spot
  • Nodule
  • Papule
  • Whitehead
  • Blackhead
• Personalized skincare recommendations using the IGA acne grading scale
• JWT-based authentication with refresh token rotation and revocation
• Microservices architecture with asynchronous inference processing

---

Architecture

The platform follows a microservices architecture, where services are independently deployed and communicate through a gateway.

Service	Tech Stack	Purpose
Auth Service	Spring Boot, Spring Security, Spring Data JPA, Hibernate, PostgreSQL	Handles JWT authentication, access tokens, refresh tokens, and token revocation
Inference Service	Python, Flask, YOLOv26 custom model, Celery, Redis	Processes uploaded images, performs acne detection, and returns structured predictions
Gateway	OpenResty (Nginx + Lua)	Reverse proxy responsible for routing, authentication validation, and CORS handling
Frontend	React, Vite, TypeScript	User interface for scans, results, authentication
Async Tasks	Celery + Redis	Background job processing for inference workloads
Database	PostgreSQL	Stores users and application metadata

---

Demo

A video demonstration of the platform is available here:

https://drive.google.com/file/d/1HCAwaoGaVFXmdRQXCtw1Cf-UNQpd55_a/view?usp=sharing

System Flow

1. User uploads a facial image from the frontend
2. Request is routed through the OpenResty gateway
3. Gateway validates the JWT token with the Auth Service
4. Image is forwarded to the Inference Service
5. Celery workers process the image asynchronously
6. The YOLOv26 model detects acne types and returns structured predictions
7. The LLM generates skincare recommendations using the IGA grading scale
8. Results are returned to the frontend

---

Note:
Running the inference service locally requires a Roboflow API key. Without it, acne detection will not function.

---

Prerequisites

• Docker
• Docker Compose
• Node.js 20+ (for frontend development)

---

Environment Variables

Create a .env file in the root directory.

ROBOFLOW_API_KEY=your_roboflow_key_here VITE_API_BASE_URL=http://localhost/api

POSTGRES_USER=postgres POSTGRES_PASSWORD=postgres

REDIS_URL=redis://redis:6379

---

Running Locally (Development)

Clone the repository:

git clone https://github.com/yourusername/dermafind.git

cd dermafind

Create the .env file with the required variables.

Start all services:

docker compose up --build

This launches:

• Auth Service
• Forum Service
• Inference Service
• OpenResty Gateway
• React Frontend
• PostgreSQL
• Redis

Access the application:

http://localhost:3000

---

Notes

• The frontend expects the API at /api (configured via VITE_API_BASE_URL)
• Large image uploads are supported (up to 16MB)
• Without a Roboflow API key, inference will fail but authentication features remain usable

---

Gateway Responsibilities

The OpenResty gateway performs the following tasks:

• Routes /api/* requests to the correct microservice
• Validates JWT tokens for protected routes
• Injects user context headers for downstream services
• Handles CORS configuration for frontend communication

---