HALL 3 - Tech Talks / Paper Presentations

Advanced data transformations accelerators, such as data lineage powered by GenAI are becoming increasingly critical, as industries prioritize data modernization, transitioning from traditional legacy systems to advanced data technologies. This paper delves into the application of LLMs in tracing the data lineage. We will share our direct experiences in addressing the issues, obstacles encountered and the effective strategies we implemented to overcome them. The proposed approach offers several advantages, including improved efficiency, accuracy, reliability and enhanced data security by keeping data knowledge within the enterprises cloud environment. The approach involves custom chunking strategies, hybrid semantic and vector search capabilities, selection of embedding model. Overall, this paper presents an efficient and secure solution for accelerating data lineage task leveraging LLMs.

The session will explore the concept of agents, delving into what they are and their potential applications. It will provide insights into when agents should and should not be used, as well as guidance on utilizing agentic frameworks effectively. Participants will learn about agentic workflows and design patterns, complemented by real-world business case studies to demonstrate their practical impact and value.

This research work presents a Generative AI framework for automated digital ad creation, addressing challenges in photorealism and product-background transitions. It proposes a three-step solution tailored to the advertising industry: (a) improved image preprocessing with Canny Edge Detection for precise inpainting, (b) optimized parameter settings for Stable Diffusion ControlNet to generate high-quality images, and (c) a customization module for templates, fonts, and text. The framework streamlines ad creation, enhances visual quality, and reduces costs, offering a practical tool for designers.

In this talk, we explore the remarkable journey of a young, dynamic team at the Indian Space Research Organisation (ISRO) as they pioneer Artificial Intelligence solutions for both space and terrestrial applications. We delve into the unique challenges of deploying AI systems in orbit—from creating robust models that function in extreme conditions to ensuring reliable decision-making under stringent resource constraints.
On Earth, these innovations fuel breakthroughs in advanced computer vision—ushering in new standards for space exploration and satellite technology. We also highlight GYAAN, an outstanding AI Assistant that reaches millions of users including students in Bihar, demonstrating how ISRO’s AI-driven achievements directly contribute to India’s growth.

Throughout this session, we reveal the step-by-step innovation process—identifying mission-critical needs, deploying generative AI for research and development, and implementing health-monitoring solutions for space missions.. We examine the successes, lessons learned, and evolving strategies that will shape ISRO’s AI-driven future. This journey of experimentation, perseverance, and creative problem-solving offers invaluable insights for any organization looking to harness AI for impact on Earth and beyond.

This research introduces an agent-driven Text-to-SQL approach using fine-tuned Small Language Models (SLMs) and Retrieval-Augmented Generation (RAG) to simplify enterprise database interactions for non-technical users. It employs specialized agents for tasks like entity recognition, query decomposition, self-correction, and explainable results. Rolled out to 700 users in a pharmaceutical company, the system achieved 87% accuracy, reduced SQL effort by 70%, improved data access by 90%, and cut costs by 50%, enabling scalable, conversational self-service analytics for business users.

This research work presents a systematic approach to case deflection, enabling customers to resolve issues independently, reducing the need for support. It develops Intellectual Capital (IC) through analyzing documentation and engineer case notes, using a predictive model with supervised learning and NLP to classify bugs for digitization. The model prioritizes high-potential bugs, optimizing workflows and resource allocation. Large language models (LLMs) automate IC document creation, improving consistency and reducing manual effort. This AI-driven process enhances decision-making, customer satisfaction, and operational efficiency.

A big agentic switch is happening globally due to AI (artificial intelligence), whereby consumers and enterprises are switching to AI agents for task completion in an autonomous way. This paradigm shift due to AI is leading to the proliferation of AI agents in enterprises requiring innovative agent network architectures for better discovery, communication and security of these agents. Here we present a novel agent network architecture where we introduce innovative concepts like agent registry, agent name server, agent text transfer protocol (ATTP), agent private cloud and agent gateway. Quite a lot of parallels have been drawn from the current world wide web (www) Internet architecture, where similar components exist to support discovery, security and communication.

This research work introduces an autonomous large multimodal (LMM)-based agent for automating software and web operations using enterprise knowledge. The agent performs repetitive web tasks, automates decision-making, and enhances precision through retrieval-augmented generation (RAG). It leverages HTML parsing to create unique identifiers and integrates self-evaluation tools for task execution. Achieving 40% accuracy on open websites and higher on internal pages, it demonstrates a 24% improvement in precision. Key metrics include execution steps, token cost, and self-reflection capabilities.

This research introduces an LLM-based agentic framework aimed at reducing Mean Time to Recovery (MTTR) during IT incidents. The conversational AIOps Incident Management bot assists on-call personnel by recommending resources, referencing similar past incidents, and suggesting collaborators. It also automates tasks like generating post-incident reports and creating Jira issues for follow-up actions. The framework integrates root cause analysis, resolution guidance, and quality monitoring to enhance efficiency, user engagement, and overall incident management in complex IT service environments.

This research explores deep learning models to predict the aging and sudden death behavior of lithium-ion batteries (LIBs), critical for automotive energy systems. Traditional models struggle with LIB aging complexity, while DL models like LSTM, GRU, TCN, and Transformers effectively forecast capacity degradation and unexpected failures. Using data from 124 cells with diverse lifetimes, advanced data practices enhanced model accuracy. Metrics like RMSE and R² validated performance, highlighting DL’s potential in improving LIB management and automotive system reliability.