NoCode Machine Learning
I'm kicking around the idea of making a no-code product that would allow you to create machine learning models without writing code (and deploy it, and use it with zapier, etc).
If you're familiar with the space it would be like a cheaper / faster nanonets.
It would allow you to do:
- image classification (is there a person in the photo)
- object detection (where is the person in the photo)
- text extraction (ocr) -> pull items from a pdf
- others?
My main concern is what people would like to use it for. What would you use it for if it existed?
Trending on Indie Hackers
I believe the idea is viable. But the problem is it's hard to implement and there is a plenty of solutions from big players like Amazon and Microsoft already exist that it will be very tough to beat.
I think on the surface that's true. But having been in this space for a while I know how expensive those APIs can get for companies. To the point where the business model is no longer viable for a lot of use cases.
Also, there is such a thing as AutoML, which is also well developed. It's not 100% "no-coding", but it's 1 step from it.
AutoML is amazing. I have a friend working on it at Google. I've been working at a computer vision startup for the past 4 years now and essentially my gripes with autoML, nanonets, etc are:
I've seen this with a lot of ML startups. It takes crazy huge scale to start getting economies of scale out of your gpu set up. I look at tensorflowjs as a way to bypass that and have it run in the browser so you could sell it for an order of magnitude cheaper.
I guess, there is always room for one more simple AI tool.
I wish you luck with this project :)
I am an ML practitioner. It means the product is not for me)
It's not clear for me. Cases you present don't need to build a model: person detection, OCR, usual NLP tasks like a topic classifier. There are ready-made models for this from Google, Amazon, Azure and other smaller companies.
There is, however, a field of custom models for custom tasks. Something like conversion probability or demand prediction for a specific chain of shops.
I love this idea. One potential use case I thought of the other day was I wanted to get data out of a PandaDoc and into a Xero invoice, and the Zapier integration and direct PandaDoc/Xero integration couldn't get it done for me. I would have loved a step in Zapier to send it to a text extraction tool to pull out some info inputted by the PandaDoc signee and send it to Xero.
Have you come across akk.io by chance? Think they are taking a similar approach.
No i haven't! akk.io looks awesome for tables. It looks like they may just stick to tables though if that works I'd bet they will expand.Interesting though. With tables, you could really just make a google sheets add on and not force the user to go to your site and upload a csv.
Hello Andrew, congratulations for such an innovative idea. However, unfortunately its already been implemented by Google. Take a look at this: https://teachablemachine.withgoogle.com
Why "unfortunately"? It means there is a market!
I’ve seen that, It’s awesome! I wouldn’t consider it a complete product though.