AI product catalog assistant
A conversational shopping assistant that helps customers find products across a 240k-item catalog using natural language, with RAG over product specs, reviews and inventory.
Client
Lumo Retail
Year
2025
Duration
9 weeks
Team
2 engineers + 1 ML specialist
The challenge.
Lumo's search relied on keyword matching, which failed when shoppers described what they wanted in their own words. Bounce rates on search were high, and merchandisers spent hours curating collections that went stale within days.
What we built.
We built a RAG-powered assistant that understands intent, filters by attributes shoppers care about (fit, material, price, sustainability), and cites the product data behind every recommendation. A merchandiser console lets the team tune the assistant's tone and pin seasonal priorities without touching code.
Results.
- ▸+31% conversion for shoppers who used the assistant
- ▸Search-to-purchase time cut from 9 minutes to under 3
- ▸Merchandiser collection work reduced by ~60% via auto-generated shortlists
- ▸Assistant answers cite sources, cutting 'is this accurate?' support tickets
The AI catalog assistant paid for itself in the first quarter. Clean code, clean handover.
Marta Silva
Head of Product, Lumo Retail
Next project
Clinic booking & records app