A searchable library of every sales call your team records

I want a searchable call library app where my sales and support teams go to find what was actually said on past calls. It is built on Google Drive, where our recordings already live, Azure AI Speech for transcription, and HubSpot for filing summaries. This is a library people open and browse, not something that runs on a schedule. Nothing should be processed unless a person asks for it.

The landing view is the recording list. Use the Google Drive List Files operation to read our call folders, and show each recording as a row with the call date, the duration, the file name, and a transcription status badge that reads either Transcribed or Not transcribed. Anything not transcribed yet gets a Transcribe button on its row. Let the user pick which Drive folders count as call folders, and let them add more folders later.

When someone clicks Transcribe, the handler pulls the audio with the Google Drive Download File Content operation and runs the Azure AI Speech Fast Transcribe Audio operation with speaker diarization turned on, so speakers come back separated. Fast Transcribe Audio is the right operation here because it handles files up to two hours and 200 MB synchronously and supports diarization. Do not use Recognize Short Audio, which caps at 60 seconds. Show clear progress while a transcription is running, and flip the badge to Transcribed when the text has been stored.

Opening a call gives a transcript reader. Show the transcript as speaker labelled blocks with a timestamp on each block so it reads like a conversation rather than one undivided block of text. Beside the transcript, show an AI generated summary of that call broken into topics discussed, objections raised, and commitments made by either side. Generate the summary once when the transcript is first stored and save it with the call, so reopening a call is instant.

The most important view is search across every stored transcript at once. A rep types something like pricing objection or a competitor name and gets every matching call back, each result showing the call name and date with the surrounding quote and the matched phrase highlighted inside it. Let them filter results by date range and by folder. Clicking a result opens that call and jumps straight to that moment in the transcript with the matching passage in view. Search has to feel instant, so it runs against the transcripts already stored in the app and never goes back out to Google Drive or Azure to answer a query.

From an open call, the user can choose to push the summary into the CRM. Find the right person with the HubSpot Search Contacts operation, let the user confirm which contact is correct, then file the summary with the HubSpot Create Note operation associated to that contact. This only ever happens when the user chooses it. Never file calls to HubSpot automatically or in bulk. Show which calls have already been filed so nobody files the same summary twice.

Two persistence rules matter throughout. Store every transcript and its summary in the app once, keyed to the Drive file, so search is instant and the text survives independently of Azure. This matters because Azure batch transcription result files expire, so the transcript text must live in the app rather than being re-fetched. And never transcribe the same recording twice. Before offering a Transcribe button, check whether a stored transcript already exists for that file, and reuse it if it does.