Mildly successful AI project #2
Incorporating the diarization work from https://github.com/sassyass/senko-transcribe, the idea was to have that help augment my meeting notes so that it could:
- read the bear note containing my notes for the meeting for better context and attendees
- combine that with the diarization results of the meeting recording to match attendees to speakers and also provide a summary
- write the results directly into a Bear note
whereas you could do diarization through a cloud provider, the idea was to keep as much local as possible to maintain privacy and any possible leak of information.
eventually through lots of iterations we were able to get it all running via goose + mcp-bear + ollama. it worked but it was slow and often lost context and you had to kick it to continue before it finished.
meanwhile, i set up an alternate workflow but through claude desktop, taking that same local transcription but having claude do all of the matching aand summarization, which was much quicker and reliable, but passed it up to the cloud.
after all of that, since work has a gemini enterprise license, i just copy my notes and the voice recording into gemini and then just paste the results into my Bear note. it’s not as automated, but it’s fast and more data-compliant.