RecoSense is a premium, full-stack e-commerce marketplace powered by an advanced hybrid analytical engine. It transforms traditional shopping into an intelligent discovery journey using RoBERTa-based sentiment analysis, LightFM hybrid filtering, and a high-performance Groq-powered AI Consultant.
- AI Consultant (Groq API): A specialized LLM-powered consultant (Llama 3 family via Groq) that provides real-time, context-aware technical advice, product comparisons, and personalized discovery paths.
- Hybrid Recommendation Engine: Leverages LightFM to combine collaborative filtering with content-based metadata for high-precision suggestions.
- Sentiment Awareness: Integrates RoBERTa (Robustly Optimized BERT Pretraining Approach) to analyze review semantics and extract deep qualitative insights.
- Smart Cold-Start: New users receive high-accuracy suggestions based on demographic profiling (age group and gender).
- Auto-Triggered Re-training: Analytical models automatically recompute recommendations as user interactions accumulate, ensuring a perpetually fresh discovery feed.
- Cinematic Design System: Modern dark-theme aesthetic featuring glassmorphism, fluid animations (Framer Motion), and a high-impact discovery-driven landing page.
- Symmetrical UI: Optimized Product Detail layouts with perfect visual alignment between imagery and technical specifications.
- Fluid Navigation: Enhanced UX with automatic Scroll-to-Top behavior on route changes and a consistent, globally-integrated footer.
- Optimized Performance: Implemented bulk-fetch API endpoints and a Synchronized Global State (UserContext) to ensure zero-latency UI updates and preference persistence.
- Unified Product Catalog: Browse thousands of electronics with high-detail cards and instant save-to-wishlist functionality.
- Smart Wishlist Dash: A centralized hub to track your favorited tech, synced in real-time across all devices.
- Activity & Reviews: Manage your community-contributed reviews, processed via sentiment-aware backend pipelines.
- Identity Management: Secure JWT-based authentication with role-based access for an administrative inventory dashboard.
| Layer | Technologies |
|---|---|
| AI / ML | Groq API (Llama 3), LightFM (Hybrid Filtering), RoBERTa (Transformers) |
| Frontend | React, framer-motion, lucide-react, React Router, Axios |
| Backend | Node.js, Express, MongoDB (Mongoose) |
| Styling | Vanilla CSS (Modern Design System with Global Tokens) |
| State | React Context (Synchronized Global Wishlist & User State) |
graph TD
A[Frontend: React + Framer Motion] --> B[Express API]
B --> C[(MongoDB: Users, Products, Reviews)]
B --> D[RetrainManager]
B --> I[Groq AI Controller]
D --> E{Interactions >= Threshold?}
E -- Yes --> F[Inference Module: RoBERTa + LightFM]
F --> G[lightfm_recs.json]
B -- GET /recommendations --> H[RecommendationController]
H -- Priority Logic --> G
H -- Fallback --> C
I -- Streaming / Chat --> J[AI Consultant UI]
- Node.js (v18+)
- MongoDB (Local instance or Atlas)
- Python 3.8+ (For LightFM/RoBERTa modules)
- Groq API Key (For the AI Consultant feature)
git clone https://github.com/ateebarman/RecoSense.git
cd RecoSense
# Install Backend
cd backend
npm install
# Install Frontend
cd ../frontend
npm installCreate a .env file in the backend/ directory:
MONGO_URI=your_mongodb_uri
PORT=5001
JWT_SECRET=your_super_secret_key
GROQ_API_KEY=your_groq_api_key
MODEL_RUN_THRESHOLD=10# In backend directory
npm run seed # Initialize test data
npm run dev # Start API server
# In frontend directory
npm run dev # Start Vite dev serverFor deep technical dives into the model architecture, see:
- Recommendation System Guide
- Quick Re-run Engine (Level 1)
- LightFM Heavy Retrain (Level 2)
- Deployment Guide
If you need to run the engine manually without using the Admin Panel:
node -e "require('./recommender/retrainManager').startModelRun()"node -e "require('./recommender/retrainManager').startRetrain()"Generated by RecoSense Engineering. Empowering the future of personalized tech discovery. ✨