Session track: Data Intelligence
Session time:
Session description:
Agentic AI systems are increasingly expected to operate across long running workflows, coordinate actions between tools, and deliver reliable responses grounded in evolving organizational knowledge. As these systems move beyond isolated chat interactions into persistent enterprise applications, memory architecture becomes a foundational design challenge that directly impacts accuracy, adaptability, and user trust. This session explores practical approaches for designing memory systems for agentic AI applications, with a focus on how graph driven knowledge models can improve contextual awareness, consistency, and decision reliability. The presentation examines common operational issues observed in production AI agents, including fragmented conversational context, inconsistent responses across sessions, and the accumulation of stale or conflicting knowledge over time. These challenges become especially visible in enterprise environments where agents interact with large volumes of interconnected data, documents, and workflows. The session highlights why treating memory as a single undifferentiated context layer often limits the effectiveness of agentic systems. The core of the session introduces a layered memory architecture that separates short term conversational context from durable long term knowledge. The discussion emphasizes how graph based representations can help model relationships between users, tasks, systems, and organizational knowledge, enabling more context aware reasoning and retrieval. Practical examples demonstrate how combining retrieval augmented generation patterns with structured graph relationships can improve continuity, reduce repetitive interactions, and support more reliable agent behavior. The session also explores governance considerations such as memory validation, knowledge freshness, auditability, and controlled promotion of information into persistent memory stores. Attendees will leave with practical design patterns and implementation guidance for building scalable agentic AI systems that remain context aware, trustworthy, and operationally reliable in real world environments.
Speaker

FourthSquare
Ajay Athitya Ramanathan is a Data and AI Engineer with a strong background in designing and delivering intelligent, data-driven solutions that support business transformation. He holds a master’s degree in Data Analytics Engineering and brings hands-on experience working across data engineering, machine learning, and generative AI systems. His work reflects a deep commitment to building scalable, responsible, and impactful AI applications that enhance productivity, decision-making, and operational efficiency. In his current role as a Data and AI Engineer at FourthSquare, Ajay focuses on architecting advanced AI-powered systems that combine large language models, automation, and cloud-native technologies. His contributions include building collaborative AI agents, implementing intelligent document processing solutions, and developing robust conversational AI platforms that improve knowledge accessibility and streamline business workflows. He has worked extensively on retrieval-augmented generation pipelines, model evaluation strategies, privacy-aware AI design, and analytics-driven monitoring to ensure reliability and trust in AI systems deployed at scale. Previously, Ajay gained valuable experience as a Data Scientist during his co-op at Boehringer Ingelheim, where he contributed to cloud-based data pipelines, smart search solutions, and natural language processing initiatives. His work involved building automated workflows, experimenting with text embeddings, and applying topic modeling techniques to better understand and classify user queries. Earlier in his career, he worked as a Software Developer at Cheeni Labs, where he helped develop recommender systems, real-time data pipelines, search optimizations, and machine learning models focused on user engagement and content discovery. Ajay’s technical expertise spans programming, machine learning frameworks, data visualization, cloud platforms, and modern AI tooling. He is particularly passionate about generative AI, applied machine learning, and the practical integration of AI into real-world systems. As a speaker, Ajay brings a practitioner’s perspective, sharing insights drawn from hands-on implementation, cross-functional collaboration, and continuous learning in rapidly evolving AI and data ecosystems.