Session 15-E
Knowledge Graphs as a requirement for AI Ready Data and LLM Accuracy
Generative AI has created new opportunities for enterprises to leverage data in powerful ways – from improving processes to creating entirely new products and services. LLMs offer unprecedented opportunity to leverage the vast stores of data residing in enterprise SQL databases, transforming how decisions are made and strategies are formed at the business-level. But concerns loom large, namely the inaccuracy of LLM responses, otherwise known as “hallucinations.” The pursuit of mitigating these hallucinations led our team to explore the integration of Knowledge Graphs as a solution. Knowledge Graphs are emerging as a crucial tool in bridging the gap between raw data and meaningful, context-rich information. In this presentation, we focus on our extensive research and development efforts to improve the accuracy of LLM-powered question-answering systems, specifically tailored for enterprise applications. We examine the role of Knowledge Graphs in providing necessary business context and semantics to fill gaps with which LLMs typically struggle. Our main finding is that responses generated by GPT-4, combined with a Knowledge Graph, were three times greater than when relying solely on data retrieved from SQL databases. This improvement underscores the potential of Knowledge Graphs in revolutionizing how enterprises need to deploy LLMs in the future. Investing in Knowledge Graph provides higher accuracy for LLM-powered question-answering systems. And ultimately, to succeed in this AI world, enterprises must treat the business context and semantics as first-class citizens in order to create AI Ready Data. This talk will cover the following emphasizes the importance of AI Ready Data by treating business context and semantics as first-class citizens such that data is accurate, explainable and governed. In the current scenario, many organizations focus on the technological aspects of AI implementation, while overlooking the critical importance of contextual understanding and semantic richness. Our research demonstrates that paying attention to these elements is not just beneficial but essential for the success of AI applications in enterprise environments, as well as driving a data-driven culture in general within an organization. technical aspects of this integration, outlining practical approaches for incorporating Knowledge Graphs into existing Retrieval Augmented Generation (RAG) architectures. We’ll provide actionable insights for CDOs and data leaders who are looking to implement this type of advanced AI capability in their organizations. We aim to equip CDOs with the knowledge and tools needed to effectively harness the power of AI with Knowledge Graphs. Ultimately, we hope to empower CDOs to move towards a future where AI is not just an automation tool, but a context-aware strategic asset for informed decision-making and competitive advantage.
Speaker
juan sequeda
Principal Scientist and Head of AI Lab, data.world
