classifai icon indicating copy to clipboard operation
classifai copied to clipboard

Site-specific custom model trained for chatbot-based search

Open jeffpaul opened this issue 2 years ago • 1 comments

Is your enhancement related to a problem? Please describe.

A nice application for intranets (and public sites that would want a similar public user-facing feature) would be for an AI service to consume all/specific site content and then provide a chatbot feature (perhaps shown on the frontend, perhaps only on a user's account page/other page, perhaps only to logged in users, perhaps as a block to be placed on an FSE template/template part) that would otherwise act akin to site search but in a natural language/chat approach. It might also be nice to be able to allow that model to consume content from other sites (e.g. others in a multisite network, other related WP instances, and otherwise fully external sites like https://developer.wordpress.org/, https://wordpress.org/documentation/, and https://make.wordpress.org/) so that output could be informed by any site/content relevant to the site's audience. Ideally the responses can be tuned to include references/links to where those responses were trained/learned from to allow someone to link out and explore that content in more detail.

Designs

TBD, but if this concept gains traction we'll probably want to think a bit about how to best handle the UX on getting a model trained on content across different sites as well as how the output/bot could look/work and where on a site might be best for defaults/options for placement.

Describe alternatives you've considered

No response

Code of Conduct

  • [X] I agree to follow this project's Code of Conduct

jeffpaul avatar Jun 01 '23 19:06 jeffpaul

@dkotter I'm going to move this to 3.5, but I think looking into perhaps how to leverage OAI Embeddings (or similar service) to consume historical site content and then provide a chatbot experience on the frontend (either in tandem with site search or perhaps as a replacement for it) would be a great item to look at next (perhaps leveraging some pre-existing custom implementations that Fueled has built elsewhere).

jeffpaul avatar Jun 10 '25 18:06 jeffpaul