Searching the transcript has the problem of missing synonyms. This can be solved by the one undeniably useful type of AI: embedding vector search. Embeddings for each line of the transcript can be calculated in advance and compared with the embeddings of the user's search. These models need only a few hundred million parameters for good results.