The uninterrupted part is the whole thing — the moment it tries to respond mid-thought you've lost the ramble. What I never solved with that setup is that the rambles are gone afterwards; the good one from three weeks ago isn't anywhere. Do you keep them, or is it purely input-and-discard?
Post-processing is underrated — it's the difference between a transcript you skim once and one you can actually feed to something else. The bit I'd want next is it persisting: cleaned up and still findable in six weeks, instead of pasted somewhere and gone. (Building in that space so I'm biased.)
That's a genuinely good hack. The other direction that works: run a cleanup pass over the raw transcript before anything downstream sees it — an LLM that's told "this is ASR output, fix the homophones and proper nouns" catches most of it, because it has the whole context instead of one word at a time. Keep the raw text around though, it's occasionally right and the cleanup is occasionally confidently wrong.
Hi, I'm Gopal.
I'm building Remindr.
An AI that remembers conversations instead of just transcribing them.
Over the next year I'll be sharing:
• Voice AI
• Engineering lessons
• Product decisions
• Startup journey
Follow if you're building AI products.
https://t.co/TFRUlj7SJv