Meet DA People #5 — Atılberk Çelebi

Hello Atılberk, can you tell us about yourself a little? What did you study and what have you been working on so far?

Hello. It is a fun story how I end up here. I originally aspired to be a comic and a cartoonist before college. Hoping that I can always fulfil my aspirations as a hobby, I decided not to spoil my good grades and enrolled to a degree in computer engineering at Koç University. Looking for an inspiration for digital visual artistry, I found myself dealing with convoluted learning algorithms and complex mathematical models. I happened to take a course on cognitive psychology, because I had read “Visual Thinking” by Rudolf Arnheim, and it had puzzled me about the human perception very much. Amazed by the obscurity of human mind, in addition to the required coursework, I overloaded and customised my curriculum with a wide range of electives in cognitive psychology, neuroscience and philosophy as well as artificial intelligence, machine learning and digital signal processing. I spent a semester in Technical University of Munich taking graduate level courses in robotics and neural engineering. I got so intrigued by the field that I completed computer engineering with a double major in electrical and electronics engineering. Beside the coursework, I also had the opportunity to assist academic research at KU AI Lab, primarily on natural language processing (NLP) problems.

After graduation, I continued my education by pursuing a masters’ degree in computer engineering at Boğaziçi University while I was working to make my living and put my knowledge into practice. Before graduation, I had the chance of spending a short period of time in the start-up ecosystem. Looking for a corporate option this time, I found a position at the R&D and Special Projects Department of Yapı Kredi Bank which gathered a great team of brilliant engineers. I contributed to the research and development of 2 major NLP projects: FOCA, financial document digitalisation and analysis automation project, and BotYap, a chatbot development platform with extensive knowledge and conversation management capabilities. Both were game changing projects for the banking industry, and they were globally awarded and praised. After my two-year tenure, I left the bank in December 2019 and joined Dogma Alares’ AI&ML team. Currently, I work as a senior machine learning engineer developing similar impactful projects for not only for banking but for various industries such as insurance, retail, logistics etc.

What should we understand when we hear AI&ML? How do you find the state of AI&ML applications in business?

A tough question. I often create analogies to better explain the broadness and vagueness of the concept. The term artificial intelligence, in my understanding, is much alike with the term art in the sense that it is hard to define and simply constrain by words. However, we practically say artificial intelligence to those tools and devices with the ability of performing tasks either just as natural as organic beings or achieving better outcomes than an average human. Let’s break this down into pieces: By “tools and devices”, I mean anything created by another agent, hence the artificialness. I could be a full-fledged autonomous humanoid robot or just a piece of digital software calculating one single number. By “performing tasks”, the emphasis is on the predefinition of the purpose, objective, target and goal. Not only the end but also the mean is important in this concept. We may desire a robot arm to move around in a factory with a soft and natural movements rather than swift and direct ones so that it can collaborate with human workers around. On the contrary, we may only want a software to automatise mundane tasks to finish them more quickly or more precisely compared to a human. therefore, I chose the word “better” to point out the use of AI does not always aim for higher accuracy, but sometimes evaluated for higher speed or standardisation or some other metric.

I am also aware that a simple calculator also falls under this definition yet no one, at least in this era, would call a calculator as artificial intelligence. Nonetheless, the state-of-the-art machine learning algorithms we employ today are only a very complicated and well-engineered versions of cascaded mathematical operations that not only model the desired behaviour but also learn to model from scratch by harvesting relevant information from existing evidence. In the past, people thought there is a simple ideal model to explain the evidence and expected that the data would fit the model; just like defining a summation algorithm and expecting every combination of numbers would give the correct result. This is almost never the case for sophisticated problems of real life. They are hard to generalise, and one may always come up with an exceptional example. Nowadays, with the discovery of more sophisticated machine learning paradigms such as deep learning and the enablement of vast amount of computation power like GPUs and TPUs, hopefully, we design algorithms which could find close-enough models out of the existing multidimensional data. This is a revolutionary change and similar to shift from the Platonic understanding of the universe to a more Aristotelian interpretation. The truth is not up in the sky of ideas yet on the ground and visible through the evidence.

The business world is still on the verge of exploiting this phenomenon. Of course, the pioneers in some industries have already utilised most of their data, collecting more pieces and employing machine learning algorithms to model some of their business functions using them. Some gathered dedicated data science and machine learning teams to tackle their problems and even to contribute academic research. Even though the others are aware of the power of data and analytics, they do not have a feasible roadmap or a strategy to implement the recipes. As far as I see and follow, there are still a huge majority of organisations they are at the very low stages of the pyramid. The pyramid of data and AI, where you start by acquiring the required data, storing and managing with proper integrated ETL flows, cleaning and preparing data, defining analytics and metrics and finally building models from the simplest to the most convoluted ones. They need to adopt a novel mindset to build up the capability from the bottom, starting from the data collection and management, they should go towards helpful and beneficial models with determination. This is certainly not an easy transformation and requires a solid strategy with an elaborate roadmap. Despite the length of the journey, it is definitely manageable with an agile methodology. Selecting and prioritising convenient and impactful cases and working as squads to discover the need of data, choosing a problem to attack, developing the models, deploying and integrating them to business operations and assuring scalability and sustainability. These are some of the things that we try to help our clients at Dogma Alares.

How did you get together with Dogma Alares? For the sake of AI/ML, how things are seen and done at Dogma Alares?

Completing two successful R&D projects at Yapı Kredi, with complete cycles from problem discovery, research, proof of concept, analysis, design, development, deployment and maintenance, I was ready for a new beginning. I was looking for some opportunities where I can contribute to similar AI&ML involving projects where I not only take part in the technical part, that is the research, the data operations and model development and deployment, but also have the grasp of the business side, discovering the need, defining the problem, setting up the goals, monitoring the outcomes and framing the governance structure. I was so lucky that you have introduced me to Erdal and Kıvanç, the founders of Dogma Alares. After a series of fruitful conversations and discussions, we realised that we want to do similar kinds of projects and aim the same sort of value. Moreover, the opportunity of joining a start-up from the early days and participating to build the company and the culture together appealed me very much. We now have a skilful and interdisciplinary AI/ML team both learning from and sharing with the others practicing service design and strategy consulting.

I believe that Dogma Alares targets a very juicy spot in consultancy and combines AI&ML right on. Climbing up the pyramid I mentioned requires a strong dedication and adopting data-oriented and analytic business manner. Wherever on the data pyramid, in parallel, the business should discover its key processes and the employer’s and the customer’s journeys along these processes so that where and how to use the power of data and machine learning can be employed. All these must be aligned with the strategy. Hence, our four-pillar vision of next-generation consulting comes into the picture. We are able to attack our clients’ needs and problems from every aspect in perfect coordination and utmost cooperation. When we develop a machine learning model, we are also aware of what we are trying to solve, which part of which process we are targeting and how the users’ experience will change with the help of the model. Furthermore, strategy perspective allows us to see the future works and assure the quality of the work by aligning various projects and initiatives together. Helping clients to solve their problems with elegant solutions and enabling them to replicate with proper organisational and technical structures pleases me very much. I hope we will grow our teams and capabilities to multiply the pleasure.