Vijoy Basu

Deep learning systems excel at pattern recognition and learning representations, but they struggle with tasks that require reasoning, planning, and algorithmic-like data manipulation. Currently, Generative AI is witnessing a pivotal shift where Data scientists are moving beyond the traditional LLM prompt-response paradigm and text-based interaction to AI systems that have much more agency. These agents employ thinking models from cognitive science to solve tasks involving long-term planning and execution. Generative agents can access real-time data and instructions from a broad range of sources, integrate various tools, have memory, derive execution plans, reflect, and orchestrate these plans. They can collaborate with other generative agents or humans and directly influence the world to solve complex tasks such as simulating human behavior, planning projects, composing music, and automating workflows. There has been substantial progress in understanding and replicating human-like problem-solving abilities in AI, with applications spread across various industries. While Generative AI is already succeeding in applications and products in the industry such as dialog systems, machine learning system augmentation, NLP tasks, and content generation - organizations often struggle to identify real-world use-cases and locate the right talent to integrate the latest tools and solve complex problems. In this talk, we will cover the current research and architecture for developing generative AI agents. We will also showcase code applications and use cases where generative AI can automate or augment workflows with human experts, and discuss how developers and data scientists can get started and leverage this paradigm in their day-to-day work.