All work
Working preview
AI product advisor
Customers describe their problem in their own words and get the right products from the shop's real catalogue.
Results
- 100% correct on a 45-question test set, run 3 times
- First answer in about 0.4 seconds
Built with
- Next.js
- TypeScript
- LLM API
- Vercel
A working preview on the real catalogue, built to show the owner before he decides.
The problem
In a shop, staff ask a few questions and point to the right oil. Online, customers had to guess between dozens of similar products, and generic chatbots invent products that don't exist.
What I built
- An advisor built on the shop's real catalogue of 41 products, in Dutch and English.
- It asks short follow-up questions, then shows 2–3 products with the reason for each and a link to buy.
- It only recommends products that exist and are in the catalogue; anything else is refused.
- A report of questions it couldn't answer, which shows the shop what customers look for and don't find.
