Hacker News Reader

Comment by halfjoking | Hacker News Reader

The quickest way to build an impressive SaaS project in 2019:

1. Find a cool machine learning project with preferably pre-trained models so you don't have to do much cleaning/moving data.

2. Get those models doing inference on a server and expose it as an API.

3. ??? (marketing/sales/business stuff)

4. Profit

View on HN

Alternative: build datasets that offers insights on X.

Examples: - Newswhip crawls news articles and uses the FB api to get share counts. Newsrooms use their data to find out what is trending

- SuperData crawls info from Twitch and YouTube to provide insights into the games that are engaging.

- Similarweb provides data into the traffic that websites are getting

- AppAnnie scrapes App rankings to provide insights into the growth and trends of apps.

- Ahrefs built a huge database of backlinks and provides insights into who is linking to your site or your competitors.

you lost me at 3.

Any examples?

There are probably a bunch on ProductHunt.

For example https://remove.bg got a lot of upvotes - it probably uses this on the backend:

https://github.com/tensorflow/models/tree/master/research/de...

So who would pay for that? I can’t imagine there’s much demand for an app to remove image backgrounds?

I think you have errors in this regard. Because in such work, Now the probability that has been created is incredible. You have been urged to investigate on some companies who belong to the same as remove.bg, for exemple: https://clippingpathindia.com

they’re selling API access, so someone who needs to do this en masse, eg if you had an app that needs avatars with a clear face pic, dating apps, etc. Perhaps researchers looking to analyse facial features who want to skip this step, law enforcement looking to clean up a bunch of selfies, media/tv companies building copy for ads, a photoshop plugin builder etc.

Building your own deep learning model is expensive and resource intensive, if it’s a solved problem it’s a great thing to outsource.

you still need to run that tensorflow model on a monster server to see performance - no ? That can be quite expensive.

That's why I put ??? as step 3. :D

I'm not the person to sell you it. That's just an example where someone might make money.

If you've got pageviews, you've potentially got revenue.

Stock/Forex predictions with multiple ML models where clients could choose which ones they would like to follow. Bonus $ for showing them the best performing model for their portfolio etc. With a simple PWA app and UI and some scalable AWS/lambda instances to handle the load and a couple of instances to keep training/predictions set going.

I feel like machine learning is to the stock exchange like regular expressions are to html: everyone who learns about the first will at some point think them to be useful for dealing with the latter.

And they will be wrong because HTML is a Chomsky Type 2 grammar (context-free grammar) and RegEx is a Chomsky Type 3 grammar (regular expression). RegEx has no memory or stack.

There are forms of ML that have memory/stack, and I would think you have to use something more complex for the stock market than "recognizing patterns". For short term trading there are definitely useful pattern recognition systems that can effectively trade.

Reply on news.ycombinator.com