Client Success Story

Helping a client enhance user experience & boost product discovery with an intelligent AI-powered search

The client

A leading bathroom and lighting company

Our client, established in 1960, is a renowned bathroom and lighting company offering a premium collection of faucets, showers, sanitary ware, flushing systems, wellness products, water heaters, and varied lighting products.

The company is recognized for its commitment to quality, durability, innovative designs, and exceptional customer service. With a keen focus on emerging lifestyle trends and sustainability, it aims to bring the best-in-class products to homes and commercial establishments.

In 2008, we started with a small web maintenance project for the client. Gradually they delegated critical tasks to our team including its website development, application development, customer support system development, and managing their India and global portal.

PROJECT REQUIREMENTS

Incorporating AI into the client's website search functionalities

With new technologies and customer expectations rising, in our recent collaboration with the client, they leveraged our AI/ML consulting services and want us to incorporate AI into their website's search capabilities. Specifically, they wanted us to build an intelligent search engine for their global eCommerce website that can understand natural language queries, determine user intent, and provide highly relevant search results for enhanced product discovery and improved user experience.

PROJECT CHALLENGES

Deploying AI-based search with advanced features and functionalities

1

Understanding English (natural) language queries

Our team had to apply NLP techniques to enable the search algorithm to comprehend natural English phrases. For example, correctly interpreting a search query like "shower under 10000" requires language processing capabilities.

2

Add predictive & assistive search features

We need to incorporate features like word completion suggestions, auto-correct of misspellings, and related keyword recommendations (based on top searches) to assist users and improve relevancy.

3

Enable dynamic & contextual search results

Our AI software developers had to ensure that the search output must change dynamically based on the search terms. For instance, results for "faucet" should differ from results for "faucet black" based on color filtering.

4

Relevancy of search results

Our biggest challenge is displaying the most useful and relevant results at the top, tailored to the user's search intent. The relevance relies heavily on our algorithms' abilities to match queries to appropriate information.

OUR SOLUTION

Replacing conventional text-based search with AI-assisted search

The client's global website is built on NopCommerce and hosted on Oracle Cloud. To replace the existing text-based search (which was keyword-specific) with an AI-based search, we came up with the following solution for the client-

Data cleaning and data pre-processing

As a part of our artificial intelligence development services, we started with the 'data preparation' process which is critical for the AI model to understand user search intent and map queries to optimal products. Our key data preparation activities include:

  • We used Exploratory Data Analysis (EDA) to identify patterns and relationships. This enables the AI model to train and learn from the users' queries and provide relevant suggestions by thoroughly analyzing search queries and buying patterns.
  • The client's product catalog had some discrepancies, to make sure that the AI training model was fed with the right data, our experts performed data cleansing and data standardization.
  • As a part of data pre-processing, we transform the cleaned data into an analysis-ready format for effectively training the AI algorithms at scale regarding search relevancy.

Feature engineering

  • We break down text from documents and product catalogs into word tokens using Python's NLTK toolkit. This allows the parsing of phrases into meaningful units for analysis.
  • Leveraging the Porter Stemmer, we reduce words to their root form. This handles different variants with a common stem to support mapping search terms.
  • With our data annotation services, we tag and label words/tokens, enabling the AI model to extract meaningful terms and relationships to understand search intent

Feature extraction & model building

  • To match search queries with the most relevant products, we used techniques like Bag Of Words (BOW), Term frequency and inverse document frequency (Tf-IDF), and Word2Vec model from genism to do embedding from generated tags.
  • Used cosine similarity and cosine distance to measure the distance and angle between the embedded vectors to enable the model to retrieve closest-matching results for a given search by comparing vector alignments.

Model deployment

  • We created a training pipeline to train our models to determine how frequently the data will be fed to the model.
  • We created an API to interact with the model.
  • We built in query processing modules to handle spell checking, autocorrecting, stemming, and other transformations on raw search terms.

Tech stack

  • Oracle Cloud
  • NopCommerce
  • MS SQL Server
  • NLP with Python NLTK
  • Computer Vision

AI/ML models

  • Bag Of Words
  • Word2Vec
  • Tf-IDF

Project outcomes

Successful implementation of AI-powered search in 3 months

In just three months, our artificial intelligence development company successfully implemented the AI-based search functionality on the client's global website.

Enhanced user experience with intelligent recommendations

The AI model provides prompt keyword suggestions after users type three letters. It guides them to relevant products even if users are unfamiliar with specific product terminology.

Comprehending natural language search queries

The search engine is able to correctly interpret English phrases like "10 liter water heater" or "single lever basin mixer in black" instead of relying on strict keywords.

Real-time spell-checking and autocorrect

The model checks spelling and automatically applies fixes by referring to the provided product catalog dataset.

CONTACT US

Need professional help with AI/ML development?

Set up a free consultation with our AI/ML consultants today! To get in touch or share your requirements, write to us at info@suntecindia.com.