Countdown to DataConnect!
00
DAY
00
HR
00
MIN
00
SEC
AI agents are often demoed as seamless, intelligent systems that can reason across documents and databases, but building the AI agent in the real world tells a different story. In this session, I will share what actually happened when I built an AI agent that interacts with PDFs and databases. I will also explain how the promising prototype quickly turned into real-world challenges, which are hallucinated answers, inconsistent query behavior, broken retrieval logic, and also edge cases that were not obvious until users got involved. This talk goes beyond how the system was built and focuses on what I learned once things started to break. I will walk through the key failure modes and the practical changes I made to improve the reliability. AI agents are not just about generating responses, they are about building systems that can act and be trusted in real workflows. The audience will leave with a clear understanding of where the systems fail and how to design them to work in the real world.
AI agents are often demoed as seamless, intelligent systems that can reason across documents and databases, but building the AI agent in the real world tells a different story. In this session, I will share what actually happened when I built an AI agent that interacts with PDFs and databases. I will also explain how the promising prototype quickly turned into real-world challenges, which are hallucinated answers, inconsistent query behavior, broken retrieval logic, and also edge cases that were not obvious until users got involved. This talk goes beyond how the system was built and focuses on what I learned once things started to break. I will walk through the key failure modes and the practical changes I made to improve the reliability. AI agents are not just about generating responses, they are about building systems that can act and be trusted in real workflows. The audience will leave with a clear understanding of where the systems fail and how to design them to work in the real world.
Hima Ganga Yarlagadda, M.S. in Management Information Systems, is a Principal AI Scientist at CVS Health specializing in Artificial Intelligence, Machine Learning, Generative AI, and Advanced Analytics. With over 11 years of experience across healthcare, banking, and insurance industries, she has led the development of large-scale AI and data science solutions that have generated significant business value through predictive modeling, fraud detection, workforce optimization, financial forecasting, and intelligent automation.Throughout her career, Hima has designed and deployed machine learning, deep learning, and large language model (LLM) solutions using technologies such as Python, Snowflake, AWS, Google Cloud Platform, LangChain, Gemini, and RAG architectures. Her work has delivered measurable outcomes, including multimillion-dollar savings through credit risk modeling, fraud detection, healthcare analytics, and workforce planning initiatives.Before her current role, Hima served as Principal Data Scientist at Discover Bank and Lead Data Science & Analytics Advisor at CVS Health, where she developed AI-powered applications, conversational agents, and predictive analytics solutions that improved operational efficiency and customer outcomes. She is a Python and SAS Certified Professional with expertise spanning machine learning, natural language processing, cloud computing, and enterprise AI strategy.Hima is passionate about advancing responsible AI adoption and helping organizations transform complex data into actionable business insights. She is a frequent speaker on topics including Generative AI, LLMs, predictive analytics, machine learning, and the future of AI-driven decision-making in enterprise environments.
