If you go the GPT route, try to make the backend flexible instead of marrying it to ChatGPT's API. Competing services (if not self hosted LLMs) could drive prices down very soon.

Also, those ideas are not mutually exclusive.