Sowmith Reddy Thukkani

Beyond the Model: Architecting Production LLM Systems in a World of Agents

Abstract:

The rapid advancement of large language models has shifted the focus of artificial intelligence from standalone models to complex, production-grade systems that integrate multiple components, tools, and workflows. Modern AI applications are no longer defined solely by model capabilities, but by how effectively these models are orchestrated within broader system architectures. This keynote explores the design and implementation of production LLM systems in an emerging landscape of agent-based intelligence.
The presentation examines key architectural patterns for building robust AI systems, including orchestration layers, retrieval-augmented pipelines, tool integration, and multi-step reasoning workflows. It highlights how agent-based approaches enable systems to dynamically plan, execute, and adapt across diverse tasks, while addressing practical challenges such as latency, scalability, observability, and reliability. Through real-world system design examples, the talk demonstrates how these architectures are transforming modern applications. It concludes by discussing the evolution toward agent-driven ecosystems, where AI systems operate as coordinated networks of models and tools to deliver intelligent, context-aware outcomes at scale.

Profile:

Sowmith Reddy Thukkani is an artificial intelligence researcher and technology professional specializing in intelligent systems, machine learning, and scalable data-driven platforms. He holds a Master’s degree in Computer Science from Eastern Illinois University and has contributed to several areas of modern computing, including conversational AI, machine learning-based anomaly detection, distributed system architectures, and intelligent data processing frameworks. His publications explore topics such as retrieval-based chatbot systems, real-time data processing architectures, visualization techniques for complex datasets, and adaptive analytical querying in distributed environments.

In addition to his scholarly contributions, Sowmith actively participates in the academic community as a peer reviewer for international conferences and scholarly venues in artificial intelligence and computing. His work focuses on designing practical AI systems that integrate conversational interfaces, intelligent data pipelines, and scalable architectures to support modern digital platforms. He is currently pursuing doctoral research in data science while also working on the development of Clonaar, an AI-driven discovery platform focused on conversational search and personalized recommendation technologies.