I wasn’t suggesting this. I was just using ChatGPT to “demo the approach”.
Assistants will be useful if you need a shortcut to providing some “conversation logic”. But this is equivalent to Completions if you are prepared to code that yourself. That’s a separate and independent choice though and if the OP is going to create UI, the end user need not know whether they employed Assistants or Completions because the power is effectively equivalent. The latter is more flexible and will potentially cost you less to operate. Assistants save you time up-front during development.
The challenge lies in the length of the stories. If these can be kept well within the size of the maximum input prompt of a model (so there’s room for the “control prompting” too), then the implementation will be much simpler.