ShopSage - Multi-AI E-commerce Agents

From search to try-on to checkout: Multi-AI agents for every shopping step. Kubernetes-powered shopping agents with virtual try-on magic

An AI-powered e-commerce assistant system built for the GKE Turns 10 Hackathon. This project enhances the Cymbal Shops online boutique (Hackathon Track) with intelligent agent capabilities using Google's Agent Development Kit (ADK).

Architecture

The system consists of five specialized agents orchestrated by a main coordinator:

Agent Details

Each agent has specialized tools and dependencies for their specific functions:

Product Finder Agent

Product Recommendation Agent

Order Placement Agent

Virtual Try-On Agent

Export Data Agent

Architecture Overview

The system follows a hierarchical agent architecture with specialized agents for different e-commerce functions:

Product Finder Agent

Purpose: Search and discover products on the Cymbal Shops website
Features:
- Real-time product search using web scraping
- Product detail retrieval with pricing and descriptions
- MCP integration for external data sources
Tools: search_products(), get_product_details()

Product Recommendation Agent

Purpose: Provide personalized product recommendations
Features:
- Category-based recommendations
- Complementary product suggestions
- Popular item recommendations
- Style and preference-based filtering
Tools: get_all_products(), recommend_products()

Order Placement Agent

Purpose: Manage shopping cart and order processing
Features:
- Add/remove items from cart
- Cart summary and total calculation
- Simulated checkout process
- Order confirmation and tracking
Tools: add_to_cart(), remove_from_cart(), view_cart(), simulate_checkout()

Virtual Try-On Agent

Purpose: Enable virtual product try-on using AI image generation
Features:
- Image processing and validation
- AI-powered virtual try-on (integration ready for nano banano API)
- Style recommendations
- Product suitability assessment
Tools: process_user_image(), generate_tryon_image(), get_style_recommendations()

Export Data Agent

Purpose: Export order data and generate professional PDF documents
Features:
- Order confirmation PDF generation
- Product details with pricing and quantities
- Shipping and payment information
- Professional formatting with tables and styling
- Artifact storage for download
Tools: export_order_to_pdf(), validate_order_data(), get_order_from_placement_agent()

Technical Details

Technologies Used

Google ADK: Agent orchestration framework
Google Vertex AI: For AI Model
BeautifulSoup4: Web scraping
Pillow: Image processing
MCP: Model Context Protocol integration
GCloud CLI: For Creating Kubernetes Cluster
kubectl CLI: For ADK deployment on GKE
Nano Banana: For Virtual TryOn
ReportLab Python Module: For Generating Order PDF

Data Sources

Primary: Cymbal Shops website (https://cymbal-shops.retail.cymbal.dev/)
Product Data: Real-time scraping from the live demo site
Cart Storage: In-memory

AI Models

LLM: Gemini 2.0 Flash for agent reasoning and orchestration
Image Generation: Gemini 2.5 Flash Image Preview for virtual try-on

Live Link (Deployed on Google Kubernetes Engine)

Visit: http://34.122.40.40/dev-ui?app=ecommerce_agent

Sample Outputs

Export Data Agent

The Export Data Agent generates professional PDF documents for order confirmations. A sample output is available:

Sample Order PDF - Demonstrates the PDF export functionality with:

Order confirmation details and tracking information
Complete product listings with prices and quantities
Shipping address and payment method information
Professional formatting with tables and branding
Total cost calculations and order summary

What I've Learned

Agent orchestration is way harder than it looks! Coordinating five AI agents taught me that it's less about prompts and more about system design. GKE deployment was a steep curve too.