danielrosehill's picture
Initial commit: Voice Note Dataset
830dfc5
raw
history blame
1.75 kB
So a very powerful agent to add to the library. For the email drafting one, I will gather these in a tag called agent plans and they can be there together. This would be a very big one if it could be pulled off. And again, I come back to the question of if there's a front load transcription that doesn't need to route everything through voice notes. For example, if the N8N form element had a voice capture node, it would solve all of this very elegantly.
So if I dictate an email and I use a transformation in voice notes, I need to say every time it's for me. It gets the names wrong all the time; it gets the style wrong. So it saves a lot of time but still a lot of nudging. System prompt to an email sending agent is like done very successfully in the past; it works very well. So the note comes in tagged, it goes to this, and the output should be basically ready to send every time. So that's not worth creating just to get to that step.
What it would be worth creating to do is a contact matching and have to be saved into and so on. So, I'm going to go ahead and start the draft probably initially until it's validated or whatever human in the loop thing they have now. In other words, if I say send this to Ronnie or an email to Ronnie or an email to Stephanie or the people that I'm sending to frequently with the Google contacts integration which exists, I imagine they have set up an MCP.
It can retrieve the person's email based on the entity match, put them in the to field, and that way I could just dictate emails that basically would be hopefully a queue of emails ready to go with one button of a push. So that would be a really cool one to try depending on if the contacts MCP is mature enough to support this.