Retrieval-Augmented Generation (RAG) has become the leading approach for equipping LLMs with enterprise-scale knowledge. However, current RAG pipelines often suffer from brittleness, relying heavily on heuristics like custom parsers, contextual chunking, and manual query optimization. In this talk, we’ll examine the challenges of today’s RAG systems and explore how advances in embedding models and rerankers are paving the way for more robust, scalable, and automated solutions, enabling higher-quality, more relevant results with less reliance on ad-hoc retrieval techniques.
