Voice of the Client: Anyline teaches smartphones to read with Watson AI
According to Statista, there are 2.5 billion smartphone users worldwide in 2019. It’s interesting that the name of the device includes the word “phone” because for many people that’s last on the list of myriad functions they use the gadget for.
Oftentimes smartphone users navigate, tell time, take pictures, pay for things, watch videos and text more often than they make calls.
Smartphones can do so much for us, but one thing they can’t do is read: they can only take a picture -- or video -- of words.
Anyline is an artificial intelligence (AI)-focused company in Austria that enables smartphones to read, interpret and process visual (analog) information in real time using a software development kit (SDK), which is easily integrated into a mobile app. Its customers are developers or agencies who are building apps for their clients.
Conducting neural network training
The SDK technology can read passports, drivers’ licenses, license plates, documents and serial numbers, as well as take meter readings.
Initially we did the training on premise in our office, but when our company grew, we knew we’d have to move to the cloud to scale. We evaluated several providers and chose IBM Cloud because of the advantages of Watson Machine Learning, which began with the ease with which we could attach its API to our training pipeline. With Watson Machine Learning we can get the pipeline up and running faster than with other cloud providers, and automate most of it, too. Additionally, we can train multiple neural networks in parallel, which saves time: in two days, we can conduct 10 trainings, whereas with workstations on premise, we could conduct one training in two days.
Putting the Anyline solution to work
We started with meter-reading and we are now an EU market leader in that discipline. There are different ways that the SDK is used. Some of our utility provider clients offer their customers a self-service option where they use their smartphone to read the meter values and transfer the data back to the utility providers, who can then bill accordingly.
There's also workforce approach where service people go from house-to-house collecting the meter data. Lastly, in the case of switching to smart meters, the technology can be used to ensure the value on the old meter is correct and stored as data to mediate risk of a client later complaining that the reading – and thus their bill – was wrong.
Additionally, 45,000 members of the Austrian police force use our solution on their smartphones. They are able to scan a driver’s license plate and obtain information about criminal history, if any. Because their smartphone solution is integrated with back-end systems, police are able to spend more time patrolling and less time filling out paperwork.
Getting to market faster
If we have a new client coming in with a unique application for our SDK, with Watson Machine Learning we don’t have to do multiple weeks of proof of concept and then multiple weeks of training before we can deliver our solution. This means we can help our clients bring a new solution to market faster.
The Anyline solution is secure by default, because all of the computing is happening directly on the smartphone. No data leaves the device, which alleviates privacy and security concerns – something that’s top-of-mind for a lot of companies today.
One of our next steps is to build a platform for our customers so that they can train their own neural networks and get their solutions up and running on the mobile phone in hours, even further reducing the time to market.