Of course, when it comes to searching the web, Google’s got you covered. However, over years of activity, organizations collect heaps of data that accessing is not always the easiest thing. Though there are companies that offer search engines for organizations, to help them combat this issue, a new Israeli startup wants to one-up them, with deep learning and some tricks from Google.
Language models that make enterprise search more useful
The Israeli startup Ask AI developed a corporate search engine, which tries to solve a known problem in every company –helping employees find where their information is, and what can be done with it. Ask AI's system generates answers and insights for organizations from their various sources of information and communication, with a focus on customer experience departments. "Today, employees in the organization are forced to go from system to system in search of answers, and often receive inaccurate links to long documents. At the same time, understanding the problems and recurring questions of customers is currently a manual and inefficient process," said Dr. Alon Talmor, the company's founder and CEO, in a conversation with Geektime.
According to Talmor, although there are very popular and powerful products like Elasticsearch for organizations to use, such products are mainly based on keyword search, while Ask AI’s development includes the use of deep learning to understand the language, context, and synonyms to significantly improve the accuracy of the search. Talmor told us that while the competitors can offer about 50% in terms of accuracy in the search, their system can reach 80-90% accuracy. "We are the first to present direct answers within the organization, similar to Google Answers. So, for example, to answer the question: 'How to add users?', instead of giving a link to the document explaining it, we actually present an answer with whatever steps are required to add a user, as seen in the document. This saves people time as they don't have to read the document themselves, especially when multiple reference sources are presented," Talmor added.
He said that the system they developed is based on the use of large language models, which they developed in-house, that learn to understand questions and target answers from any type of document or client communication– all without the need for remediation or adjustment of the documents. Talmor also points out that "Using models and studies directly from academia also raises challenges of applicability to working with real customer information, but at the moment the results and customer feedback are very promising."
From the man who sold a company to Salesforce
Ask AI was founded last year by Dr. Alon Talmor, who previously founded BlueTail, which was sold in 2012 to Salesforce. Yesterday (Tuesday) the startup came out of stealth with $9 million in seed funding. The round was led by Vertex and State of Mind Ventures, with the participation of other funds and several private investors, including GuideStar Ventures, Gefen Capital, AGP Capital, GTMFund, the Web Summit Fund, Interplay Ventures, Ran Sarig and Efi Cohen - the founders of Datorma; Guy Zipori and Or Hiltch - founders of Skyline.ai; Simon Chan's Firsthand Ventures, Omri Barzilay from Peak State Ventures, Michael Matias, and other investors. The company currently operates in a SaaS model. Ownbackup, Yotpo, Armis, Coralogix, Claroty, WalkMe and more are among its customers and development partners.