Chat-bot for the selection of wallpapers on the «Leroy Merlin» website

Basic information

01/

Customer

The Leroy Merlin company is chain of hardware hypermarkets.

Objective

The main idea is to implement the service in order to assist visitors to sort out the wide verify of wallpapers in online store.

Decision

The chat-bot with computer vision technology. It orders products whose characteristics and appearance are suitable for user requests.

Interesting fact

The service works like the Akinator game. In it the user thinks of character, and then the Genie guesses who it is by asking leading questions. In our service, the character replaced by wallpapers and Genie – by chat-bot.

Way work

02/

How the user sees it

The person wants to buy wallpaper by the Leroy Merlin website.

There is the wide verify of wallpapers on the website, and it is inconvenient for user to make a choice.

The website has chat-bot, that offers help to the user.

The buyer answers on the bot’s questions and indicates the base type (vinyl/paper), the room, for which the wallpaper is selected, the colour spectrum, the comfortable price and other important characteristics of the product.

The bot examines the photos in the store’s database and suggests the suitable examples of wallpapers.

The buyer reviews the examples, chooses the appropriate option and makes a purchase.

How it looks from the inside

Website

Widget

Chat-bot

Neural net with computer vision technology

About numbers

The Leroy Merlin online store has about 1500 titles of wallpaper

What we did:

We taught the neural net to distribute images of wallpaper into categories: photo-wallpaper, light wallpaper, wallpaper with the small pattern and so on.

What we achieved:

The neural net selects the wallpaper options that are the most suitable for the user’s description.

03/

The construction for creation multipurpose dialogue systems

The service for hosting and evaluating the computer vision models

The adaptable product search system by the photos

04/

Roman Doronin

Project manager (ruled the project, drunk coffee)

Sergey Verentsov

Backend-developer
(made the chat-bot inside the widget)

Georgiy Poputnikov

Frontend-developer
(Made the widget for the site)

Ramil Gizatullin

Data scientist
(developed the algorithm for fast image search)

Konstantin Kubrak

CV-engineer
(taught the neural net to understand wallpaper)

The MVP preparation took the only 1 week. We made it in a short time thanks to our own development platforms.

05/

It was not easy to built the architecture of the recommender system.

The bot integration with the recommendation system was the difficult process.

The first time we have solved a similar problem for the operating online store

Creating the functional and at the same time simple service – is always not the easy task

06/

«This service gives an appreciable economic effect when working with large group of goods: materials for repairs, clothing or food. The Akinator shortens the customers’s way from choosing products on the website to purchasing, furthermore, it allows the conversions increase. This is very convenient way to search for goods, especially when you don’t understand exactly what you need».

Roman Doronin

CEO EORA

Where it may be used

07/

Neural net for recognition of meter readings

Search for products by photos on the website of the online-store

Logo-plagiarism checker system

Search for furniture by photos on the website