Team project · Shipped
CompareShop
Agentic Fashion Search Assistant
A shopping copilot: describe what you want in plain language and get guided, relevant results instead of hunting through filters and categories.
Multi
source aggregation
NL → SQL
intent translation
Role
Agentic chat — the conversational assistant layer
Team
Built with a team — UI, product search, and chat as one product
The problem
Online shopping often creates more friction than convenience. Users are presented with too many options too quickly, traditional filters feel rigid and slow, and search fails as soon as someone doesn't know the exact product name — especially when they're comparing categories or trying to describe a need in plain language rather than a query.
What the product does
E-Commerce Genie acts as a shopping copilot: it understands intent through chat, converts vague shopping needs into actionable product suggestions, and narrows down relevant choices instead of forcing users through a rigid filter funnel. The aim was to make shopping feel less like searching and more like being assisted.
My contribution: the agentic chat layer
This was a team build — UI, product search and conversational assistance came together into one product. My focus was the agentic chat interface, the part that made the experience feel intelligent rather than just another product grid with a search bar on top.
Entry point
A floating chat trigger for quick access from anywhere in the product, plus a dedicated chat window for the actual shopping conversation.
Persistent state
Chat state survives across the session, so users can continue a conversation without losing context — including unique message handling and an explicit chat-reset flow.
Search-context integration
Chat interactions feed back into product discovery, so the assistant isn't a bolt-on widget sitting apart from the shopping flow — it actively guides it.
Design philosophy
The product was built around a simple idea: if users can explain what they need in their own words, the shopping experience should adapt to them — not the other way around. That shaped both the conversational assistant and the layout of the app around it.
Built in April 2025
AI tooling has moved fast since this shipped — today's models and infrastructure would let several parts of this go further: deeper personalization, richer agent memory across sessions, stronger real-time product-API integration, and better reasoning over product comparisons. The core product thesis holds regardless — reduce shopping confusion with an agentic, conversational interface instead of a filter grid.
Stack