Yeah, that's cool. Currently tangential, but conceptually not something that would be completely out of scope in the end. I'm planning to use machine translations, text-to-speech, and multi modal generative models for accessibility already. There's also an idea for baking in GPS audio tours. Obviously depends on sourcing some quality content first

When you say open source is it so you could self host it, use your "own" models, and curate your own datasets? or some other reasons? I could see a future where a lot of the project could potentially be open sourced and work with any defined geojson API.