Every transcription tool I tried failed the same way: the transcript was fine, the handoff was not.

If the buyer is a developer, they need a file they can pipe. JSON with speaker ids, start/end times, and the raw text. If the buyer is a podcast editor, they need something they can paste into a doc without cleaning quotes.

Those are two products. Shipping both as one "export" button is how you get a tool nobody finishes using.

Before you write another screen, pick the owner of the file. Then build the export around that person. Everything else is decoration.