Building Relevance Formulas with LLMs for E-commerce Search

Learn to build intelligent e-commerce search systems that combine conversational AI with traditional search functionality in this conference talk from Haystack US 2025. Explore how to create search experiences where users can express preferences through natural language while still receiving sorted results with previews and facets for manual refinement. Discover a proof-of-concept approach that represents product attributes as tensors and computes relevance scores using dot-products between product tensors and user preference tensors. Master the integration of these tensor-based calculations with traditional ranking factors like distance and price into comprehensive scoring systems. Understand how to implement LLM-powered weight tuning that allows users to conversationally adjust search parameters, enabling nuanced preference expression such as wanting both affordable and reliable products while discussing trade-offs. Gain insights into transforming traditional search boxes into intelligent chatbot interfaces that maintain the exploratory nature of e-commerce browsing while leveraging the conversational capabilities of large language models for more intuitive user interactions.