The problem with a text search is that you have to get your keywords exactly right. With LLMs you ask inexact questions like 'has the topic of X ever been disc

The problem with a text search is that you have to get your keywords exactly right. With LLMs you ask inexact questions like 'has the topic of X ever been discussed?'[1] without needing to have an exact match on X. An LLM front end which could return references to the full text seems like the best use of both. [1] For example, your query might have had X be 'crime' and the transcript would have references to multiple specific types of crime such as 'muggings', 'vandalism' etc. which a full text search isn't going to match. Further, with the LLM front-end you could refine the query to ask about violent crime etc.