If you’ve shopped on Amazon in the past few months, you might have noticed it has gotten easier to find what you’re looking for. Listings now have more images, detailed product names, and better descriptions. The website’s predictive search feature uses the listing updates to anticipate needs and suggests a list of items in real time as you type in the search bar.

The improved shopping experience is thanks to Abhishek Agrawal and his Catalog AI system. Launched in July, the tool collects information from across the Internet about products being sold on Amazon and, based on the data, updates listings to make them more detailed and organized.

Abhishek Agrawal

Employer

Amazon Web Services in Seattle

Job title

Engineering leader

**Member grade **

****Senior member

Alma maters

** **University of Allahabad in India and the Indian Statistical Institute in Kolkata

Agrawal is an engineering leader at Amazon Web Services in Seattle. An expert in AI and machine learning, the IEEE senior member worked on Microsoft’s Bing search engine before moving to Amazon. He also developed several features for Microsoft Teams, the company’s direct messaging platform.

“I’ve been working in AI for more than 20 years now,” he says. ”Seeing how much we can do with technology still amazes me.”

He shares his expertise and passion for the technology as an active member and volunteer at the IEEE Seattle Section. He organizes and hosts career development workshops that teach people to create an AI agent, which can perform tasks autonomously with minimal human oversight.

An AI career inspired by a computer

Agrawal was born and raised in Chirgaon, a remote village in Uttar Pradesh, India. When he was growing up, no one in Chirgaon had a computer. His family owned a pharmacy, which Agrawal was expected to join after he graduated from high school. Instead, his uncle and older brother encouraged him to attend college and find his own passion.

He enjoyed mathematics and physics, and he decided to pursue a bachelor’s degree in statistics at the University of Allahabad. After graduating in 1996, he pursued a master’s degree** in statistics, statistical quality control, and operations research **at the Indian Statistical Institute in Kolkata.

While at the ISI, he saw a computer for the first time in the laboratory of Nikhil R. Pal, an electronics and communication sciences professor. Pal worked on identifying abnormal clumps of cells in mammogram images using thefuzzy c-means model, a data-clustering technique employing a machine learning algorithm.

Agrawal earned his master’s degree in 1998. He was so inspired by Pal’s work, he says, that he stayed on at the university to earn a second master’s degree, in computer science.

After graduating in 2001, he joined Novell as a senior software engineer working out of its Bengaluru office in India. He helped develop iFolder, a storage platform that allows users across different computers to back up, access, and manage their files.

After four years, Agrawal left Novell to join Microsoft as a software design engineer, working at the company’s Hyderabad campus in India. He was part of a team developing a system to upgrade Microsoft’s software from XP to Vista.

Two years later, he was transferred to the group developing Bing, a replacement for Microsoft’s Live Search, which had been launched in 2006.

Improving Microsoft’s search engine

Live Search had a traffic rate of less than 2 percent and struggled to keep up with Google’s faster-paced, more user-friendly system, Agrawal says. He was tasked with improving search results but, Agrawal says, he and his team didn’t have enough user search data to train their machine learning model.

Data for location-specific queries, such as nearby coffee shops or restaurants, was especially important, he says.

To overcome those challenges, the team used deterministic algorithms to create a more structured search. Such algorithms give the same answers for any query that uses the same specific terms. The process gets results by taking keywords—such as locations, dates, and prices—and finding them on webpages. To help the search engine understand what users need, Agrawal developed a query clarifier that asked them to refine their search. The machine learning tool then ranked the results from most to least relevant.

To test new features before they were launched, Agrawal and his team built an online A/B experimentation platform. Controlled tests were completed on different versions of the products, and the platform ran performance and user engagement metrics, then it produced a scorecard to show changes for updated features.

Bing launched in 2009 and is now the world’s second-largest search engine, according toBlack Raven.

Throughout his 10 years of working on the system, Agrawal upgraded it. He also worked with the advertising department to improve Microsoft’s services on Bing. Ads relevant to a person’s search are listed among the search results.

“The work seems easy,” Agrawal says, “but behind every search engine are hundreds of engineers powering ads, query formulations, rankings, relevance, and location detection.”

Testing products before launch

Agrawal was promoted to software development manager in 2010. Five years later he was transferred to Microsoft’s Seattle offices. At the time, the company was deploying new features for existing platforms without first testing them to ensure effectiveness. Instead, they measured their performance after release, Agrawal says, and that was wreaking havoc.

He proposed using his online A/B experimentation platform on all Microsoft products, not just Bing. His supervisor approved the idea. In six months Agrawal and his team modified the tool for company-wide use. Thanks to the platform, he says, Microsoft was able to smoothly deploy up-to-date products to users.

After another two years, he was promoted to principal engineering manager of Microsoft Teams, which was facing issues with user experience, he says.

“Many employees received between 50 and 100 messages a day—which became overwhelming for them,” Agrawal says. To lessen the stress, he led a team that developed the system’s first machine learning feature: Trending. It prioritized the five most important messages users should focus on. Agrawal also led the launch of incorporating emoji reactions, screen sharing, and video calls for Teams.

In 2020 he was ready for new experiences, he says, and he left Microsoft to join Amazon as an engineering leader.

Improved Amazon shopping

Agrawal led an Amazon team that manually collected information about products from the company’s retail catalog to create a glossary. The data, which included product dimensions, color, and manufacturer, was used to standardize the language found in product descriptions to keep listings more consistent.

That is especially important when it comes to third-party sellers, he notes. Sellers listing a product had been entering as much or as little information as they wanted. Agrawal built a system that automatically suggests language from the glossary as the seller types.

He also developed an AI algorithm that utilizes the glossary’s terminology to refine search results based on what a user types into the search bar. When a shopper types “red mixer,” for example, the algorithm lists products under the search bar that match the description. The shopper can then click on a product from the list.

In 2023 the retailer’s catalog became too large for Agrawal and his team to collect information manually, so they built an AI tool to do it for them. It became the foundation for Amazon’s Catalog AI system.

After gathering information about products from around the Web, Catalog AI uses large language models to update Amazon listings with missing information, correct errors, and rewrite titles and product specifications to make them clearer for the customer, Agrawal says.

The company expects the AI tool to increase sales this year by US $7.5 billion, according to a Fox News report in July.

Finding purpose at IEEE

Since Agrawal joined IEEE last December, he has been elevated to senior member and has become an active volunteer.

“Being part of IEEE has opened doors for collaboration, mentorship, and professional growth,” he says. “IEEE has strengthened both my technical knowledge and my leadership skills, helping me progress in my career.”

Agrawal is the social media chair of the IEEE Seattle Section. He is also vice chair of the IEEE Computational Intelligence Society.

He was a workshop cochair for the IEEE New Era AI World Leaders Summit, which was held from 5 to 7 December in Seattle. The event brought together government and industry leaders, as well as researchers and innovators working on AI, intelligent devices, unmanned aerial vehicles, and similar technologies. They explored how new tools could be used in cybersecurity, the medical field, and national disaster rescue missions.

Agrawal says he stays up to date on cutting-edge technologies by peer-reviewing 15 IEEE journals.

“The organization plays a very important role in bringing authenticity to anything that it does,” he says. “If a journal article has the IEEE logo, you can believe that it was thoroughly and diligently reviewed.”