Talk 1: Optimising Data for LLMs: Strategies for effective RAG
Jordan Moshcovitis
Data preparation not only enhances the accuracy of Retrieval-Augmented Generation (RAG) systems for Large Language Models (LLMs) but is also effective in improving their output relevance. This presentation will examine strategies that boost the semantic integrity and efficiency of data retrieval systems supporting LLMs. We will explore the progression from traditional RAG techniques to advanced methods such as propositional and semantic chunking, emphasising the role of thorough data preprocessing. Additionally, the discussion will include the deployment o tools that facilitate semantic enhancements and the deserialization of complex document structures, which collectively refine the efficacy of RAG applications.
About the presenter: Jordan is an engineer and researcher with a formal background in physics and mathematics. His current focus is on developing application-layer services for large language models, specifically information retrieval techniques like retrieval augmented generation (RAG) and LLM agents. Jordan has demonstrated his expertise through a variety of projects, from constructing bionic eye prototypes to deploying advanced ML models for uncertainty quantification in physical systems. His career is marked by a passion for applying quantitative methods in diverse settings, tackling complex problems and an passion for innovation in technology. With understanding of both the theoretical underpinnings and practical applications of LLMs, Jordan works to extend the capabilities of LLMs to transform complex data into actionable insights and bridge the gap between theoretical research and practical, scalable applications in AI.
