E-commerceOpenAIPython

AI Shopping Assistant

Conversational product finder for an appliance store

Shoppers think in tasks: “I need a quiet air conditioner for a 20 m² bedroom”. Catalogue filters ask for exact specs, and anyone who does not know them calls a manager or leaves the site. We built an AI assistant into an online store of climate and home appliances that helps shoppers pick products in a plain conversation. It takes prices, stock and specs only from the store's backend and never makes anything up.

The task

The store needed product selection by need, not one more filter. The assistant had to:

Take a request as free text, with no required fields
Understand the intent and extract product parameters from the text
Ask follow-up questions when data is missing, instead of long forms
Search products through the store's existing API
Keep the context of the conversation
Answer common questions about delivery, payment and warranty
Hand the shopper over to a person when it cannot solve the request

An add-on, not a new platform

We did not build a separate AI platform or rewrite the store: the assistant works as an add-on to the existing backend. The backend stays the source of data and business logic, and a new AI Service in Python and LangChain runs the dialogue and the work with the model. Every price or status in the chat comes from the store's API at the moment of the answer; the model only understands the request and phrases the text.

A request in the shopper's own words

The assistant lives right in the catalogue: a button in the corner offers to pick a model for anyone unsure which capacity to choose. The shopper writes the way they think, and the assistant shows which parameters it understood and asks one or two follow-up questions with ready-made reply buttons. A short conversation instead of a form with a dozen fields.

Catalogue and assistant button

A regular catalogue page with filters; the assistant button with a hint offers to pick a model for shoppers who do not know the specs. AIRO is a demo store, models and prices are made up.

1 of 3

Products with real prices and stock

Once it has enough parameters, the assistant searches the store's API and shows products as cards: model, key specs, price, stock status and a link to the product page. The model is not allowed to generate prices, specs, stock or delivery terms, so the price in the chat equals the price in the cart, and the assistant never recommends what is out of stock.

Product picks

Three models for a sunny room and a budget under 25,000 UAH, with prices and status from the backend, and a note on which one is the quietest.

1 of 2

Common questions and a hand-over to a person

The assistant answers questions about delivery, payment and warranty from the store's rules and brings the shopper back to the selection. It does not try to solve project requests, such as fitting out an office with installation: the conversation goes to a manager together with its history, and the manager continues it in the same window.

Delivery, payment, warranty

An answer to a common question based on the store's rules, and an offer to get back to choosing a model.

1 of 3

What we built

Chat widget on the website with the history of the current dialogue
AI Service: a separate module between the widget, the model and the backend that runs the dialogue scenarios
Intent recognition and parameter extraction from free text
Parameter search through the store's backend API
Recommendations as product cards linking to the product page
Answers to common questions about delivery, payment and warranty
Rules for the AI: what it may do, which API methods it may call, how it handles personal data
Hand-over to support when a request goes beyond the assistant

The result

Shoppers who do not know the exact specs now get from their task to a product page without filters and without a manager.

The store completes the selection where it used to lose the shopper
The assistant closes part of the enquiries, so managers work on the complex ones
Request logs show what is missing in the range and in product descriptions
The approach carries over to any catalogue with an API

With our technical process

Our clients' ideas rock

E-COMMERCE LEADS

Automationn8nCRM IntegrationE-commerce

A unified system for processing leads from all channels — validation, CRM deal creation, distribution among managers, and instant client response.

View details

CONSULTING LEADS

AutomationAIn8nLead Qualification

Automated initial client interaction powered by n8n and AI — filtering non-target inquiries and passing only qualified leads to managers.

View details

AUTOFLOW

AutomationAI AssistantBookingAuto Service

An AI assistant answers on every channel, collects the car, problem and urgency, checks the live schedule and turns each request into a ready-to-confirm booking.

View details

CAR IMPORT

AutomationAI AssistantAuction MatchingCar Import

An AI assistant answers car enquiries on every channel, collects budget, make, year, engine and market, matches live US & Korean auction cars and hands the manager a qualified buyer.

View details

ClinicPilot

Medical practicesClaudeLaravel

AI administrator for private clinics: booking, reminders, pre-visit questions and follow-ups in Telegram, WhatsApp and on the website, with one admin panel for the team.

View details

Homeland

Logistics & TMSreactlaravel

A dispatcher hub and driver mobile app that turn check calls into live data: driver statuses, automatic location sharing, rich chat with document scanning, and two-way TMS sync — multi-carrier by design.

View details