Like much of the tech world, UruIT has been increasingly working with machine learning and “smart” application concepts. The Uruguayan app development and technology company sees the use cases for these innovations continuing to explode in the years to come.
“It is difficult to see a domain where it couldn’t be applied nowadays,” said Waldemar López of UruIT in a recent interview with Cognitive Business News.
Photo: Waldemer Lopez of UruIT is the lead author of a new e-book on machine learning. (Credit: UruIT)
In addition to his day-to-day work building technology for the firm, López is the lead author of UruIT’s recently published e-book, “The Business Executive’s Guide to Smart Applications.”
The goal for the company, which has its headquarters in Montevideo and has offices in Miami, Los Angeles, and Medellín (Colombia), is to offer practical advice that is often missing in a space where resources tend to be created for either technical wizards or complete novices — but not both.
This e-book looks to find some middle ground, offering real-world use cases that a non-tech-expert CEO can use to make decisions as well as some application-building pointers that can benefit even the most savvy programmer.
To learn more about the approach and UruIT’s overall outlook on machine learning and artificial intelligence, we recently spoke with López.
“It’s important for us to show how to move from the algorithms to a deployed application.” – Waldemar López or UruIT
Jared Wade: What is the general theme of the book and who is it tailored toward?
Waldemar López: Our idea was to introduce the development of intelligent applications and the potential of these kind of applications and then relate this to the Internet of Things and Big Data. We wanted to give some examples, explore the basic concepts, and, mainly, introduce this all to people who work in the software industry but do not have previous knowledge in the subject.
With the e-book, we divided it into a few different sections. Some of them are more for business people and other sections are more technical. We talk about some tools, languages, or techniques that usually are used in this kind of project. In this first e-book, we want to summarize all of these topics in such a way that it could be easy to read for anyone.
Jared Wade: There seems to be a lot of stuff in the industry that is highly technical for the people who are creating algorithms and the underlying technology. And then there is a lot of super basic stuff. But that middle ground seems to be lacking. So I imagine this will be pretty well received for people who are actually trying to decide how to incorporate this in their company from a business perspective.
Waldemar López: Yes, exactly. There is a lot of work done by scientists and those kind of people. But, in this case, we want to focus on how to leverage these techniques, these tools, these algorithms. We are looking at how to begin applying them to the organization or to information systems. At UruIT, we are focused on web applications and mobile applications. So how can these new technologies be applied in those kind of systems? How can we move forward from traditional ways of developed systems?
For example, in this guide, you are going to see that we talk about a traditional system like a news webpage. How can we improve it in some smart way to personalize the content of the news for each kind of user? How can this be related to the recommendation systems on how to improve and personalize the user experience? These are the kind of questions we answer in the e-book.
Jared Wade: I noticed that the book includes some definitions upfront, which I think is very useful in this space. There are so many different companies, and the terminology is not universal across the board. When you guys are talking about machine learning and predictive analysis, what is your outlook on that? Do you have a real-world use case that helps illustrate this?
Waldemar López: Another important thing for us is to give the big picture and not only the algorithms that are in the background. So we talk about the pipeline and the workflow that we generally need to work with when we start, for instance, a machine learning project.
We included an interesting image in the book about all the different steps — starting from collecting the data — to help people understand how to begin a project. Or to help people understand what kind of models can be applied on that data — not only discussing the algorithms for this model, but also how we are going to deploy this model and how it’s maintained. There are lot of things — not only algorithms — that are important when we want to address this kind of project in a software company.
“This year, and probably next year, we are going to see some important improvement related to customer relationship management.” – Waldemar López or UruIT
We also recently worked on an article about a practical machine learning use case that is very related to the e-book. In it, we move through the pipeline and show, through Python examples, how to build an application from scratch. It is a very interesting application in this case that works with the football, or soccer, video game FIFA 18.
A guy collected all the player data through a web-scraping model written in Python. So we downloaded this data and, over it, we created different machine learning models, and we ended up by building a basic web application. The results are displayed in an interactive way where the user can play with the model, see what prediction it makes, and see the real values versus the prediction error.
We go through all the Python from scratch — collecting the data, how the data is studied, and how different models compare. And at the end we talk about building an application. This step is generally missed when we read different posts about machine learning that concentrate on the algorithms itself. It’s important for us to show how to move from the algorithms to a deployed application.
Jared Wade: Is it able to predict that Uruguay is going to win the World Cup?
Waldemar López: I guess that would be very easy — and we can hard code that prediction!
Jared Wade: When you talk to customers, what are you guys hearing? What are the things that they are the most interested in bringing to market now?
Waldemar López: I personally like a lot of the current research that is being done in text analytics. I am currently working on some projects to extract the semantics of the text. So, for example, this is not only useful when building a chatbot for support, but it also can be used to identify related customer cases that had a similar final solution for the issue.
Also, related to this, we are looking at how to extract information that is currently in a non-structured format, seeing how we can extract the relevant features of this information and give recommendations, for example. Recommendation systems are very-well-known systems with some classical techniques. But I think that, right now, when you see some deep learning techniques related to text analytics, we can extract relevant information from the descriptions of the product. And you can collect all the relevant information without having a guy that is trying to classify each product in different classes to train a model.
I see a very interesting new approach there of collecting all that information, all that semantic information — in a smart way. So I think that, this year, and probably next year, we are going to see some important improvement in that area related to customer relationship management.
Besides information management, we’ve seen other areas where machine learning could have an impactful influence for companies. So far, we’ve had several clients that are using algorithms to generate matches between users, vendors, consumers, employers, and employees, among others.
Additionally, in the telecommunications and banking sectors there are good opportunities to use machine learning to predict customer churn and many other issues. Or companies from other sectors can use the strategy of clustering to segment their market and make better marketing and business decisions. And the list goes on.
“There are good opportunities to use machine learning to predict customer churn. Or companies can use the strategy of clustering to segment their market and make better marketing and business decisions.” – Waldemar López of UruIT
Jared Wade: For 2018, other than things we’ve talked about, are there any other big emerging applications in this space? Or perhaps applications that are going to gain much more adoption as far as bringing together smart applications of machine learning?
Waldemar López: In the future, AI will generate a big impact in building a better, more equal society in different ways.
One example: The digital information collected during the last few years can be exploited now by smart agents, and, soon, “machine reading” will extract knowledge from these sources. With that, these agents will be able to support virtual assistants to help doctors by suggesting diagnostics or acting as teachers recommending content or answering questions. The latest cognitive computing developments, like image or speech recognition, will also improve the interaction with these agents.
AI combined with the Internet of Things will collect tons of data in smart cities, and this will contribute to create safer environments by using facial recognition to identify criminals or detecting hidden patterns from sensor data to predict important events. Self-driving cars are something that we see everyday in the news. And it is something that is happening today. At the enterprise level, AI will be used to automate different process, from recruiting to virtual assistants helping customers.
I think that we are going to see a ton of news in this year related with automation where machine learning is applied everywhere. It is difficult to see a domain where it couldn’t be applied nowadays.
This interview has been lightly edited for clarity and space.