Invited Speakers

Contemporary generative AI systems demonstrate remarkable capabilities across creative and technical domains. They compose symphonies, debug complex software, and engage in sophisticated philosophical discourse. Yet these same systems exhibit significant limitations in humanity’s most pressing applications. They struggle with personalized mental health interventions, environmental crisis prediction, and accessible trustworthy legal services for populations.

This performance disparity illuminates a critical knowledge gap in current AI development. Recent research has identified a fundamental “discrepancy between a model’s internal understanding and the knowledge required for coherent, personalized conversations” in multi-turn dialogues. While these models excel in well-documented domains with abundant training data, they lack the specialized knowledge representations and contextual grounding required for high-stakes social applications. The resulting systems often produce outputs that are misaligned with user needs, potentially unsafe for vulnerable populations, and prone to reproducing training data rather than adapting to sensitive contexts. This talk examines the fundamental challenges underlying this knowledge gap and explores Neurosymbolic AI as a method for developing grounded and instructible AI systems that can effectively serve society’s critical needs.

Companies are rapidly adopting AI solutions like Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) to enhance employee productivity, business efficiency, and customer experience. However, generic pre-trained models often underperform compared to models fine-tuned or augmented with specific enterprise data. To unlock the full potential of AI, ingesting data from various applications and databases into a centralized Lakehouse is crucial. This talk will explore the challenges of building robust data pipelines for this purpose. We’ll delve into the complexities of handling diverse APIs, incremental ingestion, application behavior, rate limits, and data access controls, with specific considerations for unstructured data. We’ll also discuss strategies to overcome these hurdles and establish a foundation for scalable and efficient AI deployments within an organization.