When it comes to AI / ML programming it’s not necessarily about coding unique APIs and selling them from the top. More often than not you’ll be in the second tier – integrating commercial APIs into your existing platforms, to interact with your current code, interact with each other and of course interact with your user-base.
This is what my company does. We have never written a smidgen of unique AI code. But what we have been around the block with is integrating various machine learning APIs into our existing code. This means an abundance of research and lots of demos. It is, of course, different strokes for different folks but our dev team is an experienced one in the realm of artificial intelligence source code.
So I listed out 7 different useful ML categories of product that your project or business will likely encounter and I got our dev. guys to list me their favorites. Now when it comes to your turn you know exactly what to head for.
1. What’s Your Favourite Chatbot API?
BotFramework was their answer.
Ok so maybe it is a bit of a cop out going with Microsoft’s offering but the big boys do some things right. The two aspects that the guys really like are the versatility and ease in which the bot integrates cross-platform. And ‘adding smarts’ which is Microsoft’s cognitive offering.
2. What Is Your Favorite Voice Automaton API?
wit.ai was their answer.
Now you could accuse me/them of hypocrisy here. We have sort of already covered language processing with Microsoft’s ‘one size fits all’ bot. But we are specifically talking about a voice interface here. Something that wit.ai does really well and is very easy to implement.
3. What Is Your Favourite NLP API?
Google speech-to-text was MY answer.
As you would expect there was an abundance arguing over this question. For a start, there are so many different options out there. I actually decided to answer this one myself. I am not a coder but I have thoroughly demoed over 20 different speech APIs and no matter what you read, Google is king here – in my humble opinion at least. I am basing this purely on language accuracy, but let’s be honest – if you are implementing a speech-to-text feature into your app or website this is going to be your number 1 priority.
4. What Is Your Favourite Visual Recognition API?
I thought TensorFlow was a shoe in here just because of Google’s visual reach. I certainly doubt it’s far away from the others but this isn’t about me! These are developer’s answers for developers.
First off we have Clarifai. Easy to implement, milliseconds to analyze visual data, and a nice little touch of giving you the perceived accuracy score next to the picture. Also great for video & motion detection.
CloudSight is probably the most accurate API on the market in terms of scene classification and image conceptualisation and this is why it is great code to implement. BUT and there is a big but here, it is not all AI! The process of classification can up to 25 seconds, so what is going on here you may ask? Well, humans are going on here! Tagging and captioning away whenever their ML model doesn’t quite cut it. But if a slight time-lag and cost doesn’t deter you then I suggest checking this out.
NB Notable mention to VisualGraph who got acquired by Pinterest in 2014 when most of us barely knew AI image recognition even existed! Although they are strictly in-house now.
5. What Is Your Favourite Time Series / Real Time Bidding API?
Probably unequivocally because there isn’t too much else to choose from. And I’ll be the first to admit we are getting a bit vertical here. But using machine learning for RTB is an extremely interesting and profitable consideration if you or your company pushes a lot of money into ad spend.
Forget about your proprietary product for a second, and have a quick look at ML APIs that you can implement to improve your company’s internal operations. This one could be a god-send.
6. What Is Your Favourite Data Science API?
Lots to choose from. We all like H20.ai.
There really is much of a muchness here. And no-one could really justifiably, objectively answer why it is better than TensorFlow, Watson, bigML or the tens of other very, very capable predictive analytics software.
What we will say is check out their product ‘deep water’. Now, this is cool. So already GPUs are being employed to run huge amounts of parallel workloads unlike CPUs or the like. This, of course, is hugely helpful to ML analytics as a ton of unstructured data can be analyzed concordantly. However, what then happens if you want to combine different GPU technologies for speed processing of the biggest of big data. This is when you turn to H20.
7. What Is Your Favourite Facial Recognition API?
We are a fan of emovu.com. But we are the first to admit that we haven’t tried many in this space.
Emovu is one of the big guns in the space. Desktop and mobile SDK and an API. There are also a ton of features to choose from. They have been in the industry for a while, have a good user-base and are obviously well funded because of their complete and in-depth product offering. They also offer a free demo too so don’t just take our word for it!
One that I personally like the look of but haven’t given a go so far nviso.ch. The reason it appeals to me is their focus on 3D imaging. This is always going to make your results more accurate especially in a live video environment. The word of warning that comes with this tech, so I’ve heard, is that if you are looking for a walk in the park, implementation-wise, this is probably not the API for you.
And that’s that. 7 considerations, that as a developer looking into the machine learning space, you will likely encounter. Happy hunting and enjoy playing around with some very cool software.