I gave it a paragraph "hiring" GPT-4 and explaining our mission. Then a sentence overview of our two projects (api and app) as well as a description of similar apps and how we're different. Then I have about 12 tickets written in plaintext like this:
Issue ID: GD-012
Type: Task
Title: [api] Migrate to TRPC from express/rest
Assignee: John
Status: In Progress
--Description Beginning--
--Description End--
Because the outputs are really long, here's an example of my interactions with Dave (the name I gave to GPT-4). There's also emails Dave has created to fictional stakeholders but that's too high up. Right now the problem I'm having is that sometimes Dave can remember the issue tracker state but when trying to output (so that I can store it) it can't produce a long enough output (now, it worked before when I had 6 tickets). If I were to cram everything into 1 prompt then it would probably work, but a better solution is to use langchain and a document loader for issues.
I believe that if I was able to vectorize the codebase so that if could search for relavent portions of code (at work a lot of our tickets were "update this endpoint to handle case X, we did it to this other endpoint 6 months ago (link to PR)) and have a proper store of issues then it could be powerful.
This is awesome , will you be able to share the initial prompt. When I tried the same prompt to create epic, it’s missing multiple stories on GPT-4. Like validating OTP, Resending OTP and Logging for audit it did generate the other 3 though. Not sure why. But yours is spot on.
