AI-Powered E-Commerce Recommendation Engine

Overview

Built entirely by Abdul Sattar for the BIT Final Year Project at Islamia University of Bahawalpur. Full storefront plus RecommendationEngine combining user–user cosine collaborative filtering and TF-IDF content similarity.

Problem

Generic product catalogs do not personalize. The FYP goal was a working store with measurable hybrid recommendation quality.

Solution

Log weighted interactions (view/search/like/cart/purchase), build collaborative and content models, merge hybrid scores, persist Recommendation rows, and surface them in the storefront.

Architecture

User action → UserInteraction → RecommendationEngine (collaborative + TF-IDF content → 60/40 hybrid) → Recommendation table → storefront product lists.

Implementation

Django 4.2 models (Category, Product, UserInteraction, Recommendation, Order, Feedback, UserProfile), Scikit-learn cosine/TF-IDF, management command generate_recommendations, Bootstrap 5 UI, Cloudinary-ready media.

Results

Working AI shop with hybrid recommendations validated at 85%+ precision (CV).

Lessons learned

Interaction weighting and hybrid blending matter more than a single algorithm for cold-start and niche catalogs.

Key modules

• store models
• RecommendationEngine
• generate_recommendations command

Features

• Full e-commerce catalog, cart/order, and auth flows
• Collaborative filtering (user–user cosine similarity)
• Content-based TF-IDF on product text
• Hybrid merge (~60% collaborative / 40% content)
• Weighted interactions: view/search/like/cart/purchase
• Admin + recommendation generation command

My responsibilities

• Solo design and implementation of store + ML recommender
• Feature engineering for interaction weights
• Validation and deployment path (PythonAnywhere / MySQL per CV)

Outcomes

• 85%+ recommendation precision on validation