* Your users won't have enough data
* Your users will have bad data
* Your users will have sensitive data that they can't upload
* Your model would need to be tweaked for best outcomes
Sampling the background for the correct pitch would be enough
I think that's just the current world, friend.
if you want to raise vc money, the thing they care the most is the potential market size. this sounds like a niche market, so you will have to target big enterprise users, not hobbyists. big enterprise users can easily hire a data scientist team to build models, unless your solution is cheaper and flexible enough to handle their needs well. You will also need to take care of data security. And you need to have sales or you need to do sales.
if this is a bootstrap/indie hacker idea that doesn't need vc money, I still don't think it is a good idea. personally, I won't choose anything that requires a GPU infrastructure , it is too costly, and hence risky.