Sharing how I’m prototyping assistant coaches

I just wanted to share how we’ve been prototyping an assistant using the new interface.

We’ve built up a massive community of designers and established guides on things like career mentoring and portfolio building.

We’ve been wanting to train a GPT, to provide service’s coaching when we can’t always be there.

I didn’t want to write a massive long prompt so I setup a process that enables a user to trigger “prompt journeys”.

First the prompt tells GPT on its first response to give the user a code word to access the journeys. For our assistant it’s “I need help with my portfolio”.

GPT responds with a set of common services.

For most of the service’s, GPT will send the user a PDF file that’s a “workshop canvas” for a guided workshop and offer to step them through it over chat.

Throughout the sesion GPT refers first to a “GPT Portfolio Coach Guide” document for a range of questions and scenarios that it can use to provide feedback and prompts at any time.

It’s not perfect yet but I’m definitely trying to improve it. It would be great if I could set “preferred prompt suggestions”, I like forward to being able to train the model with NLP.