Alex Allan is the CTO & Co-Founder of Kortical, an AI platform company that helps enterprises develop AI practices and tools to scale and automate aspects of their businesses such as transport cost optimization, communication handling, and survey data labeling. Kortical’s customers range from agriculture, manufacturing, and media, to finance, compliance and healthcare.
What is your background and what made you start Kortical?
I was always a fan of AI in science fiction and subsequently was lucky enough to do a degree in AI back, before it was cool in 2006. This lead to winning a PhD scholarship in Data Science in 2010. While finishing the PhD, I started an AI consultancy, and then a product startup around AI for contact centers. This was where I met Andy, my co-founder, and when we first began working on automating a lot of the machine learning challenges we came across.
We quickly realised that contact centers were just a small fragment of a much larger opportunity, and thus Kortical was born. The DNA of the technologies really arose from of Andy’s background in enterprise cloud engineering and my background in AI.
What is Kortical’s mission?
Our objective is to reduce the time, cost and risk of going from zero to production with AI. We use cloud scale, machine learning methods and intuitive new ways of working to empower data scientists to build, explain, deploy and maintain machine learning solutions from POC to production.
We automate the model creation for novices, while still allowing the expert data scientist full control over every aspect of the process.
Our corporate mission is to ensure that the benefits of AI don’t stay in the hands of the few and that everyone should benefit from the next industrial revolution.
Who are Kortical’s customers and how do you create value for them? What challenges do they face, or will they face in the future, that Kortical can help them overcome?
Our customers range from agriculture, manufacturing and media, to finance, compliance and healthcare. A recurring challenge we see is that they understand the power of AI, but have skill shortages and so lack the power to get from ideation to proof point to production.
Walk us through Kortical’s products, and what problems they solve. Which product are you currently focusing the most on? Which product do you think is creating most value for your clients, and why?
Kortical is a platform which acts as a horizontal AI engine upon which a huge number of apps and services that require AI can be built. For example, in one bank we were able to use Kortical to power a back office automation app, which read email in an inbox and automatically forwarded it to the correct banking team. By building on top of the Kortical platform, we were able to go from 0 to a pilot, and we could demonstrate in a live sandbox environment in less than 3 weeks.
Often our customers choose to use Kortical to power their own AI applications, but sometimes we work with them to help deliver the application on top of the core platform.
A flagship project we are heavily involved with, at the moment, is with the NHS – where we are powering blood logistics planning across the UK. Highly accurate models are created by the platform daily, which predict demand and supply of various blood products across many different regions. This is then fed into a sophisticated next best action system, which attempts to minimise waste and transport costs, while improving availability. This will be the NHS’s first live AI system.
From our customers perspective, success is measured in speed to value. Zappi, a market research automation firm and one of our customers, was able to achieve best in market, superhuman, results automatically labeling survey data, then get the resulting Kortical model live, within their own system, in under 2 months.
How do you measure the performance of your product?
There are many open data science challenges out there on sites, such as Kaggle – we are able to automatically rank among some of the best data scientists in the world, across a huge number of different ML problems. On a number of challenges, we can get global best scores with default settings.
From our customers perspective, success is measured in speed to value. Zappi, a market research automation firm and one of our customers, was able to achieve best in market, superhuman, results automatically labeling survey data, then get the resulting Kortical model live, within their own system, in under 2 months.
Tell us about the Kortical team.
Built on a foundation of a small team, but great minds, who are crazy about AI and ensuring that we make it accessible to businesses and it is delivered in a platform that is easy to use.
There are many skills required to deliver great models that work in business at scale and we find the thrill in delivering a solution that takes away the pain of delivering brilliant AI models at enterprise scale.
How many data scientists or machine learning engineers do you have in your team?
Four – it is a small, but mighty team, that delivers a lot due to the platform doing a lot of the heavy lifting. We have a much larger number of developers who are gradually becoming data scientists through virtue of working on the platform.
We practice any form of supervised learning and our suite of models cover all the family favourites, such as deep neural networks, gradient boosting machines, linear models and NLP vector space transformations.
Which AI technologies are you experts on and are you leveraging the most?
We practice any form of supervised learning and our suite of models cover all the family favourites, such as deep neural networks, gradient boosting machines, linear models and NLP vector space transformations.
We use best in class open source solutions, such as XGBoost and Tensorflow under the hood. Our secret sauce is the layer above where we have combined the best in class model/architecture search techniques to allow the platform to rapidly design novel, bespoke and highly accurate configurations of these model building blocks.
What’s unique about Kortical versus other automated machine learning is our ability to explicitly override the search process, where appropriate. For example, if you know you want a neural network with 100 hidden units on the first layer, but don’t know how many should be on the 2nd layer or what the activation function should be, you can configure Kortical to search for just those parameters. This override can be applied to any part of the model building process and so allows expert data science knowledge to work hand in hand with powerful model search and tuning.
The ability to explain any model created on the platform is also a key part of our technology stack, and helps bring AI to more regulated sectors who may fear the black box.
What skills do you look for when hiring data scientists?
Smart people, who are passionate about the field and can map the problem to the data available. Communication and statistics are must-have core competencies.
What are the biggest challenges that you are currently facing at Kortical?
From a business point of view, we were slightly ahead of the game when we started out and it was quite painful having to do a lot of POCs to educate customers of the value of AI. Now people are much more savvy and immediately get the benefits of the platform.
From a tech point of view, many of the open source packages most data scientists use don’t ‘cut the mustard’ when you try and do things at scale and in production. We had to rebuild a number of common components to make them work at the level of performance we require.
What are Kortical’s biggest achievements in the last 12 months?
Winning Innovate UK Grant with the NHS that will create the first NHS AI project put into production nationwide. Spearheading an internal transformation project within Deloitte to automate their tax practice with our platform. Winning a tech sprint with the FCA by building a working app to identify vulnerable pensioners through data in 36 hours. Going head to head with Google and AWS in a Schroders datathon and winning by creating a churn model which could save hundreds of millions in less than 2 weeks.
We also got listed as one of the top 15 ML companies to watch in Europe by Forbes.
Which AI-related technology trend do you think will have the biggest impact in your industry in the coming years?
Online learning – essentially automating the model building and productionisation process so that models get smarter over time and can react to changes in data. This will be super powerful as AI becomes business as usual. This is a key reason you need a platform such as Kortical, as without automation model updates because prohibitively costly to do manually.
Explainability of complex and interdependent models at scale will have a huge impact too.
Dr. Alexander Allan – co-founder and CTO of Kortical
Alex did one of the first modern AI degrees available, starting a BSc in Artificial Intelligence and Cybernetics in 2006. Alex went on to win a PhD scholarship in Data Science, due to being the only 3rd year student to get a paper published in an international conference. Alex went on to win 3 best paper awards for his work in embedding spaces, and started an AI consultancy in the 3rd year of his PhD.
By the time Alex finished university, he was providing data science consultancy and building machine learning solutions for a number of well known brands, and used this capital to found an AI startup targeting lead optimisation in call centers. Alex asked his friend Andy to join him and together they built out the technology to drive the solution. This was the first iteration of what would become the Kortical platform.
Alex and Andy soon realised that the technology they had built had potential far beyond call centers, and thus Kortical was born (after buying out the previous startup).
