TransactAI: Natural Language E-commerce with AI Agents

Architecture-Diagram

Inspiration

The future of e-commerce lies in natural conversation. We envisioned a world where shopping online feels as natural as talking to a knowledgeable sales assistant. With Google's GKE turning 10 and the rise of AI agents, we saw an opportunity to demonstrate how modern cloud infrastructure and AI can transform the e-commerce experience.

What it does

TransactAI creates a seamless bridge between natural language and e-commerce: Users can ask about products in plain English ("show me what's available") Place orders naturally ("I want to buy 2 sunglasses") Get instant product information ("tell me about the hairdryer") Complete transactions seamlessly with automated payment processing Receive real-time order confirmation and tracking details All this while working with a real e-commerce backend (Google's Online Boutique), demonstrating practical, production-ready capabilities.

How we built it

We created a microservices architecture leveraging Google Cloud's best practices: Orchestrator AI Agent: Built with Streamlit and Google's Gemini Pro, handles user interactions and coordinates between services Order Agent: Implements MCP (Model Context Protocol) for service discovery and product management Payment AI Agent: Uses Mock UP A2A (Agent-to-Agent) protocol for secure payment processing Infrastructure: Deployed on Google Kubernetes Engine (GKE) Integrated with Google's Online Boutique microservices demo FastAPI for robust API endpoints Python for backend logic

Challenges we ran into

Online Boutique Integration:
- Navigating HTML responses instead of REST APIs
- Managing session cookies for cart/checkout flow
- Handling form data with specific field requirements
AI Integration:
- Fine-tuning prompts for consistent JSON outputs
- Handling safety settings and content filtering
Microservices Communication:
- Implementing A2A and MCP protocols from scratch
- Managing state across distributed services
- Ensuring robust error handling

Accomplishments that we're proud of

Seamless User Experience: Created a natural language interface that actually works with a real e-commerce system
Technical Innovation: Successfully implemented two different agent communication protocols (A2A and MCP)
Google Cloud Integration: Full integration with GKE and Online Boutique
AI Implementation: Effective use of Gemini Pro for natural language understanding
Error Handling: Robust system that gracefully handles edge cases and provides clear feedback

What we learned

Google Cloud Expertise:
GKE deployment strategies
Microservices architecture best practices
Online Boutique's internal architecture
AI Development:
- Gemini Pro capabilities and limitations
- Prompt engineering for structured outputs
- AI safety and content filtering
System Design:
- Agent communication protocols
- Session management in distributed systems
- Error handling in AI applications

What's next for TransactAI: Natural Language E-commerce with AI Agents

Implement payment AI agent communication with orchestration layer using A2A protocol.
Enhanced AI Capabilities:
- Product recommendations based on conversation history
- Price negotiation and deal finding
- Multi-language support
- Voice interface integration
Technical Improvements:
- Kubernetes auto-scaling implementation
- Enhanced caching for better performance
- More payment methods and gateways
Business Features:
- Customer preference learning
- Personalized shopping experiences
- Integration with multiple e-commerce platforms
- Analytics dashboard for merchants