Searchable customer interview library from your Zoom calls

I want an app my product and research team opens whenever they need to know what customers actually said, instead of re-watching call recordings. It is a customer interview library built on our Zoom cloud recordings. The transcripts and highlights it collects are the app's own stored data, so it gets more valuable every time somebody transcribes another call.

The main screen lists our Zoom cloud recordings from the last 90 days. Use Zoom's List all recordings operation for the list, and Get meeting recordings to reach the individual files for a session so we can find its audio file. Default the list to audio files only, and let me clear that filter if I ever need to see the other file types. Each row shows the meeting topic, the date, the duration, and a clear transcribed or not transcribed status.

Rows that are not transcribed yet get a Transcribe button. Pressing it sends that session's audio to Cartesia's Transcribe Audio (STT) operation, which runs the Ink Whisper model, and saves the returned transcript with the app, keyed to the Zoom recording. Once a transcript is stored it is never regenerated: reopening the recording, searching it, or coming back next month all read the saved copy. Show progress while a transcription is running, and a clear error state if it fails, so someone can retry that one recording without disturbing the rest.

Opening a recording shows the full transcript beside the meeting details (topic, date, duration, host, participants). The transcript should be comfortable to read for a long session and easy to scroll.

A search box at the top of the app runs across every stored transcript at once. Someone types pricing or onboarding and sees every moment it came up, each hit showing the surrounding sentence or two with the term highlighted, plus the interview name and date next to it. Clicking a hit opens that recording's transcript at that passage. Keep the search literal and fast, a plain case-insensitive substring match across the stored transcripts. Do not promise semantic or meaning-based search.

From an open transcript I can select a passage and save it as a highlight. Saving asks me to tag it with a theme such as pricing, churn risk, or feature request. A highlight stores the quoted text, the theme tag, the recording it came from, the date, and who saved it.

A themes board groups every saved quote by tag, so I can open pricing and read all fourteen quotes from across our interviews with the interview name and date on each one. From a theme, a Send to Notion action writes that theme and its quotes into a research page: use Notion's Create a Page operation for the page itself, then Append Block Children to fill it with the theme name, each quote, and the interview and date each quote came from.

Keep the theme tags in one shared editable list that everybody picks from, so nobody invents five names for the same insight. Let someone add, rename, or remove a tag in that list, and have renames carry through to the quotes already tagged with it. Show who transcribed each recording and who tagged each highlight, so the team can divide the work and see what is left to do.

Two things to get right on the data path. Zoom recording download links need our Zoom credentials attached, so fetch the audio bytes server-side before handing them to Cartesia rather than passing a bare URL along. And treat the stored transcripts and highlights as the app's own durable records rather than something recomputed on load, since the whole point is that this library compounds.