Michael Elkin, the Head of AI & Data Integration at GigaSpaces, was invited as a keynote speaker at the Future of AI conference, held last Wednesday in Tel Aviv. Speaking amongst executives from IBM, Microsoft and promising startups in the GenAI domain, Michael’s presentation focused on the challenge of bringing generative AI for structured data to a human-level interaction. To his point, GenAI tools today run into difficulty when faced with the task of analyzing structured data: such as the company’s internal sales records, inventory management, order payments, or marketing analytics. As a result, even the best LLMs today reach accuracy levels of no more than 55%, when querying structured data.

Michael Elkin, GigaSpaces. Photo: Niv Kantor

Mr. Elkin explained how GigaSpaces decided to address this challenge with an innovative new approach that provides LLMs with the “missing context” about the nature of the structured data. “When AI models look at your company’s internal records it sees a neatly organized cupboard full of drawers, but it really can’t tell what each drawer is containing or the rationale of sorting the company data in the way it’s currently stored. This is exactly why current LLMs fail to provide accurate responses to queries on structure data. We’re on a mission to change that”.