Marianne Mullan

At ITV, researchers and business users face a significant bottleneck: unlocking and synthesising insights from massive repositories of historical research data to determine what audiences want next. To solve this, ITV developed "Content Genie," a custom RAG-based system to intelligently search and synthesise research on demand. Out-of-the-box RAG tools often fail to deliver anything more than a hyper-search tool, so we needed something custom. This session therefore provides a technical, no-frills deep dive into a real-world example of ITV’s custom Retrieval-Augmented Generation (RAG) architecture. This session covers the challenges of data ingestion across diverse formats, embedding, and summarisation strategies. Crucially, it highlights how ITV moved beyond basic semantic search by implementing self-querying, metadata extraction, and advanced prompt engineering to mitigate hallucinations and drive agentic behaviour. Beyond the technical build, the session also shares the practical realities of embedding a tool like this within a business, including fostering AI literacy and utilising collaborative, human-in-the-loop design to navigate workforce hesitations.