Gift Galore

MERN-based ML-Enhanced E-Commerce Platform for Gifting Products

Comment

Inspiration

We looked at the current market of online gifting platforms and saw a universal problem: static catalogs that fail to adapt to individual preferences. The experience felt impersonal and outdated. We were inspired to build something smarter — an interactive platform that truly understands its users, recommends gifts intelligently, and provides real-time AI assistance to make gift-finding a joy, not a chore.

Gift Galore leverages machine learning, RAG-powered chatbots, and full-stack development to deliver a seamless, data-driven gifting experience. It transforms a standard online store into an intelligent, user-focused marketplace, offering personalized recommendations, actionable insights, and a smoother, more enjoyable shopping journey.

What it does

Gift Galore is a MERN-based, ML-enhanced full-stack e-commerce platform with:

Core E-Commerce Features

  • User Authentication & Authorization: Ensures secure login and role-based access, protecting user data and providing admins with full control over platform management.
  • Product Catalog: Streamlines product discovery with intuitive browsing, searching, and filtering capabilities.
  • Shopping Cart: Simplifies the checkout process with a persistent cart that lets users add, remove, and update items effortlessly.
  • Order Management: Enables users to place, track, and manage orders in real-time, improving satisfaction.
  • Payment Integration: Guarantees secure, seamless transactions via Stripe, ensuring user trust and convenience.
  • Automated Email Notifications: Enhances user engagement with automated, timely updates on order status and promotions.

AI & ML Enhancements

  • AI Chatbot (Groq’s LLaMA 3 + RAG): Provides instant, personalized gift suggestions and customer support, helping users find the perfect gift faster and more efficiently.
  • Market Basket Analysis (Apriori): Boosts average order value and helps users discover complementary gifts effortlessly by intelligently recommending frequently bought-together items.
  • Sentiment analysis: Summarizes product reviews, allowing users to quickly understand customer feedback and make confident purchase decisions.
  • Collaborative Filtering: Suggests products purchased by similar users, delivering highly relevant recommendations and improving overall shopping satisfaction.

Analytics Dashboard

  • Performance Tracking: Visualize sales trends (daily, monthly, yearly), revenue growth, and average order value helping admins make data-driven business decisions.
  • Product Insights: Identify top products, analyze category-wise sales distribution, and get sales forecasts to optimize inventory.
  • Customer Intelligence: Track and reduce customer churn, and perform advanced RFM analysis to segment users for targeted marketing.
  • Operational Efficiency: Monitor order statuses with interactive charts to streamline fulfillment.

UI/UX

  • Fully Responsive, mobile-first design.
  • Clean, modern, and intuitive interface.
  • Best-seller tags based on product performance metrics.

How we built it

  • Frontend: React.js, CSS3, HTML5 for a dynamic, responsive, and modern user interface.
  • Backend: Node.js, Express.js for a high-performance, scalable, and event-driven server.
  • Database: MongoDB with Mongoose for flexible schema design and rapid development.
  • Authentication: JSON Web Tokens (JWT) for stateless, secure, and verifiable user sessions.
  • Payments: Stripe API for its robust security and developer-friendly payment processing.
  • Media Storage: Cloudinary for optimized, cloud-based image and media management.
  • Machine Learning: Python with Pandas for its powerful data science ecosystem.
  • Chatbot: Groq’s LLaMA 3 for its fast inference speeds, integrated with Node backend
  • Other Tools: Node-Cron for scheduled tasks

Challenges We Ran Into

  • Secure Payments

    • Handling Stripe webhooks and keeping order statuses synchronized was complex.
    • Solution: Implemented robust webhook handling and order state validation to ensure accurate, reliable payments.

  • AI + Web Integration

    • Running Python ML pipelines alongside a Node.js API required careful orchestration.
    • Solution: Exposed ML functionality as a microservice with REST endpoints, allowing seamless integration and scalability.

  • Chatbot Context Management

    • Needed to balance generic LLM knowledge with app-specific FAQs for relevant responses.
    • Solution: Built a custom domain documentation dataset, ensuring the chatbot provides accurate, context-aware suggestions.

  • UI/UX for Data-Heavy Features

    • Making analytics dashboards intuitive and visually appealing was challenging.
    • Solution: Designed clean, interactive visualizations that allow admins to quickly interpret insights and act on data.

Accomplishments We're Proud Of

  • Engineered a fully functional AI-enhanced e-commerce platform from end-to-end, combining ML, RAG chatbots, and full-stack development.
  • Successfully integrated machine learning models with the MERN stack, enabling real-time, intelligent recommendations.
  • Deployed a sophisticated RAG-powered chatbot that delivers personalized, context-aware support, improving user engagement and satisfaction.
  • Implemented two distinct recommendation systems to enhance discovery and sales:
    • Apriori-based Market Basket Analysis for suggesting frequently bought-together items.
    • Collaborative Filtering for personalized “Recommended for You” suggestions.
  • Designed a comprehensive visual analytics dashboard that hat translates raw data into actionable business insights.

What We Learned

  • Designing scalable MongoDB schemas to support real-world e-commerce applications.
  • Building JWT-protected, role-based routes for secure user authentication and authorization.
  • Integrating machine learning models with a MERN stack to provide real-time, intelligent features.
  • Applying the Apriori Algorithm to transaction data for actionable recommendation systems.
  • Implementing User-Based Collaborative Filtering using cosine similarity for personalized product suggestions.
  • Using NLP techniques to summarize customer sentiment from product reviews, enhancing decision-making and user trust.
  • Developing a RAG-powered chatbot for context-aware, personalized customer support.

A Glimpse of the Math Behind Recommendations

Apriori (Association Rules)

Used for Market Basket Analysis to recommend frequently bought-together items.

  • Support of an itemset X: $$ \text{Support}(X) = \frac{\text{Number of transactions containing } X}{\text{Total number of transactions}} $$

  • Confidence of a rule X → Y: $$ \text{Confidence}(X \to Y) = \frac{\text{Support}(X \cup Y)}{\text{Support}(X)} $$

  • Lift of a rule: $$ \text{Lift}(X \to Y) = \frac{\text{Confidence}(X \to Y)}{\text{Support}(Y)} $$

Impact: Helps users discover complementary gifts, boosting sales and enhancing shopping experience.

Collaborative Filtering (Cosine Similarity)

Generates personalized product recommendations based on similar users’ purchase history.

  • Compute cosine similarity between users (u) and (v): $$ \text{Similarity}(u , v) = \frac{\text{u . v}}{\text{||u|| x ||v||}} $$
  • (u . v) = dot product of overlapping purchases
  • (||u||) and (||v||) = vector magnitudes

Impact: Suggests items the target user hasn’t bought but are popular among similar users, improving engagement and conversions.

Retrieval-Augmented Generation (RAG) Workflow

Powering the context-aware AI chatbot:

  1. Load & Chunk: Split app documentation and FAQs into ~500-char chunks.
  2. Embed: Each chunk is converted into a vector embedding using @xenova/transformers and stored.
  3. Query Embedding: Convert user messages into embeddings.
  4. Retrieve: Selectthe top 3 most relevant chunks using cosine similarity.
  5. Construct Prompt: Merge chunks with instructions for concise Markdown answers.
  6. Generate Response: Use Groq LLM (llama-3.1-8b-instant) to provide context-aware, personalized responses.

Impact: This transforms a generic LLM into a specialist expert on our platform, ensuring the chatbot gives accurate, app-specific advice, improving user satisfaction and reducing support workload.

RFM Segmentation Workflow

Used for customer segmentation and targeted recommendations:

  1. Aggregate Orders: Compute each user’s last purchase date, total orders, and total spending.
  2. Score RFM: Assign 1–3 scores for:
    • Recency: More recent purchases → higher score
    • Frequency: Higher number of orders → higher score
    • Monetary: Higher spending → higher score
  3. Assign Segments: Classify users into:
    • VIP, Loyal, Churned, Bargain Hunter, New Customer, No Orders

Impact: Allows hyper-personalized marketing and targeted promotions, improving retention, revenue, and user engagement.

What's Next for Gift Galore

  • Voice-Enabled Chatbot: Implement a voice-enabled chatbot to pioneer conversational commerce, making gift discovery faster and more accessible.
  • Deep Learning-Based Recommendations: Transition from traditional ML to Deep Learning models for even more nuanced and accurate personalization.
  • AR/VR Previews: Allow users to visualize gifts in their own space before purchase, enhancing purchase confidence and reducing returns.

Built With

Share this project:

Updates