How teomesrkhani/full-stack-painting-recommender-platform works
Project type and purpose
• This is a full-stack MERN (MongoDB, Express, React, Node.js) web application augmented with two Python-based microservices (a Flask image-generation service using PyTorch and a Chroma-based recommendation service).
• Purpose: Let users browse and save paintings, generate AI-driven new paintings via WikiArt prompts, and get personalized painting recommendations.
File structure
Top-level:
• client/ – React frontend (Vite)
• server/ – Node.js/Express backend plus two Python microservices
server/ subfolders:
• db/ – MongoDB connection helper
• routes/ – Express API routes (record.js)
• generate_painting/ – Flask service for AI painting generation
• recommend/ – Python services for embeddings and Chromadb recommendation
• server.js – Express app entry point
Key insights from README
• Frontend: React components under client/src
• Backend: REST API via Express, connects to MongoDB Atlas
• Two separate Python processes:
• Flask on port 5001 running generate_painting.py
• Python Chroma recommendation scripts
• LocalStorage fallback in client for unsaved state
Main components
a. Client (React SPA)
– Components: Home, Content, GenAI, Paintings, SavedPaintings
– Utilities: db.js (API client), localStorageDb.js, userSession.js
b. API Gateway (Node/Express server.js + routes)
– Endpoints:
• POST /record – log user interactions
• GET /paintings, /recommend – fetch data
c. MongoDB Atlas
d. Python Flask Microservice (generate_painting)
– Endpoint: /generate → runs PyTorch model using prompt_template.txt
e. Python Recommendation Service (ChromaDB)
– chroma_service.py serves recommendation queries via embedding lookup
– generate_embeddings.py prepares embeddings from WikiArt dataset
f. External services
– WikiArt API for painting metadata
– PyTorch model weights
Component relationships & data flows
1. User clicks “Generate AI Painting” in React → React calls Express API (e.g., POST /generate).
2. Express forwards request to Flask microservice (HTTP call to localhost:5001).
3. Flask returns generated image URL/base64 → Express returns to React.
4. User browses paintings → React calls Express (GET /paintings) → Express queries MongoDB Atlas → returns list.
5. For recommendations → React calls Express (GET /recommend) → Express invokes Python Chroma service or reads from local chroma.sqlite3 → returns recommended painting IDs → React displays results.
6. User actions (save, view) → React stores locally and via Express POST /record → Express writes to MongoDB.
Architectural patterns & design principles
• Separation of concerns:
• UI layer (React SPA)
• API layer (Express)
• Data layer (MongoDB, SQLite/Chroma for embeddings)
• AI/compute layer (Flask+PyTorch, Chroma recommendation scripts)
• Microservices for heavy ML workloads, keeping Node process lightweight
• LocalStorage fallback for offline UX