Co located ▼

Locations ▼

Request your invite

Partnership opportunities

Save $100 on your pass

Call to action

Your text goes here. Insert your content, thoughts, or information in this space.

Button

Back to speakers

Shubham

Maurya

Senior Data Scientist, Product Development

Mastercard

Shubham Maurya is a researcher and innovator in Artificial Intelligence, currently serving as a Senior Data Scientist at Mastercard’s Product Development. With nearly eight years of experience in machine learning, MLOps, and LLMOps, he specializes in developing scalable models and generative AI solutions. His work spans from foundational research to real-world applications, where he plays a crucial role in projects that enable near real-time inference and distributed model training. Additionally, Shubham has made significant contributions to AI optimization, notably with the "Owl Search Algorithm," a novel, nature-inspired heuristic approach to hyper-parameter tuning for global optimization. Currently, Shubham is focused on developing agentic solutions and researching advanced techniques for context document retrieval, with the goal of enhancing AI’s ability to process and utilize information more effectively.

Button

11 September 2024 10:00 - 10:30

Next-gen context retrieval: Navigating through the RAG's challenges in generative AI

Large Language Models (LLMs) power cutting-edge Generative AI applications, yet their effectiveness is heavily dependent on the efficient retrieval of relevant context. While Retrieval-Augmented Generation (RAG) has emerged as a powerful alternative to fine-tuning, real-world implementations reveal critical challenges: hallucinations, retrieval latency, lack of adaptability, and context drift. In this talk, we dissect the limitations of both fine-tuning and traditional RAG-based methods, highlighting real-world bottlenecks in context retrieval. We introduce Adaptive Context Retrieval (ACR), a dynamic approach that refines retrieval mechanisms based on evolving user queries, document structures, and contextual dependencies. By leveraging agentic workflows, hybrid retrieval techniques, and reinforcement learning, ACR promises to bridge the gap between static retrieval and real-time adaptability.