This is my prompt pipeline for arpa.chat . There are basically both hidden, public view only, and public editable prompts that append to the pipeline.
For creative tasks like coding, generative media, etc. There is also a prompt enhancer that pulls gemini api to take whatever the user is writting and make it better and tailored for the underlying AI’s logic. Eg. if you are prompting Veo, and ask “a dog running”, the enhancer will take that and create a json structure veo would expect and fill the blanks like, what kind of dog, where is it running, background details, camera details and more, making it easy to generate high quality outputs from basic prompts.
In addition, the customization itself can be streamlined in a user friendly way. Most AI platforms allow for prompt customization, but they expect that everyone knows how to prompt like a PRO. We all know that is hardly the case, even with experienced users.
For example, you can give toggle buttons or selection boxes to users, to customize their settings, which would not just change ui/ux, but also append relevant prompts to the pipeline. Eg. if user selects he wants the AI to respond with short responses, he only clicks short, and in the backend you append a detailed prompt that teaches the agent to respond shortly.