Ask HN: Lean Validation of AI Startups

Y	Hacker News new \| ask \| show \| jobs

2 points by mxmpawn 2499 days ago

Hi, I'm Software Engineer building AI projects for almost four years. I've been wanting to start selling AI products to businesses but I don't know how to validate an idea quickly, without spending time/money developing a demo.

For example, I think that a software for the construction industry that uses cameras to detect if the workers are using the proper security equipment would be a good and useful product. I can ask and then find someone from a potential company that could be interested in this kind of product.

The problem is what to show to this potential client in order to get a deal. I could develop a demo, but I think it's too expensive to build a good model. If I want to test another usecase (for example, house floor identification for real estate) I'd then have to build a model for that.

Do you have experience on this industry doing something similar to what I describe? What are you doing to validate quickly if an idea is good or not?

Maybe a good way would be to start working as a consultant/freelancer, validating demand for a product this way. But that is a slower path and you need time to work as a consultant/freelancer.

1 comments

streetcat1 2499 days ago

So you first issue is data. For any problem you would probably need a sizable training dataset.

Hence, your idea should come after the data, not before.

In this case, I would suggest going to companies that already have data (for example support tickets), which they do not use.

After you have data, I would use auto ml.

link

mxmpawn 2499 days ago

But you would need a company to give you data with nothing visible on return. Do you think a company would do that?

link

streetcat1 2499 days ago

Yes. For example, if your problem is important for construction companies, than they will give you data to solve it (otherwise, how would they solve it). In return, they will get to use your models.

In any rate, unless you have a dataset of your own, the ML part does not matter.

I think that any ML project today, it more of a solution engineering type project, and not a product.

link