I designed and built a live web application that pairs a real-time product recommendation engine with a genuine AI conversational agent. I own every piece of it myself, from the backend and API integration to the scoring logic and the interface. It is a working proof of what AI-driven personalization actually takes to ship, not just plan.

My role & impact

• Full-stack ownership: Designed, built, and deployed every part of the application myself, backend, front end, and infrastructure.
• Live AI integration: Built a PHP backend that calls the Anthropic API directly with my own key, including a full tool-calling loop, not a canned chatbot response.
• Real-time personalization: Built a content-based behavioral scoring model that re-ranks the entire product grid instantly as a visitor browses, no page reload.
• Security-conscious deployment: Isolated the API key and config outside the public web root, with a basic request cap to protect against abuse.

What I led

• Architecture: Decided how the recommendation engine and the AI agent would share state, so a conversation and a click produce the same effect on the page.
• Tool-calling design: Built a boost_interest tool the AI agent can call mid-conversation to adjust live rankings on a shopper's behalf.
• Graceful degradation: Designed the page so the recommendation engine keeps working even if the AI concierge endpoint is ever unreachable.
• Deployment: Moved the project from a hosted demo environment to my own subdomain, with my own server-side API integration.

Selected outcomes

• A fully working, live demonstration of agentic AI taking action within a page, not just answering questions.
• A real-time personalization model that requires no training pipeline and updates instantly.
• A self-contained, portable build that runs on ordinary PHP hosting, no specialized infrastructure required.