You do realise that's going to be a metric fuckton harder than targeting neurotypicals, right? Like, bordering on impossible.
The clue is in the D of ND. To put it another way, let's forget the entire spectrum of ND for a second and focus on ASD.
You not only need to train your AI on every possible interaction quirk an ASD person can have, such as trigger phrases to avoid and jobs they absolutely will not be able to do, you need it to be adaptable such that it can be useful to high functioning ASDers who can mask, to low functiiners that may not be able to leave the house but can maybe do some light computing work, and everywhere in between. And you need it to be able to detect which one it is dealing with.
That's an impossible task, because the exact combination of issues, quirks, triggers, etc, are often very rare, if not completely unique.
But surely the AI can learn what the quirks of an individual are, right? Nope. AI learning relies on large datasets to do its work. Datasets that will not exist for all except the most common of issues and quirks. The most an AI can do is avoid a given topic when asked.
Now extrapolate that to the entire ND community. Good luck.