Welcome to another installment of Arbol's Expert Insights series!

Igor Stankevich is the Managing Director of Climate Risk at Arbol, leading a team focused on Machine Learning models and risk development. Integral to his work are advanced APIs and internal and external pricing tools which are supported by a robust cloud infrastructure.

‍

Q: You have over 20 years of software development experience and a background in AI, finance, and commodities. How do you believe your professional experience has influenced your role at Arbol?

It all started with probability theory, which I found intellectually stimulating to study in university. It’s one thing to know probability concepts, and quite another to experience them in real life; playing poker with a few friends gave me that opportunity. Moreover, although my first master’s was not related to programming, I started coding while studying for my degree, and continued to do so at the small commodities trading firm where I used to work. Eventually, I started exploring strategies for automation which led me to Cognitive Science and AI. At one point, I got sidetracked working on Computer Vision – a sub field of AI – because it was a hot field back in 2012-2015.

These experiences helped shape my engineering, coding and management skills, which I used to develop a research and modeling framework that became widely adopted at the company I used to work for. At Arbol, I’ve drawn on these past experiences as I’ve worked with my team here to develop a risk framework that allows the company to systematically evaluate risks and price the products we sell.

Q: Caching is one of the toughest problems in computer science. Can you explain, in layperson's terms, how you overcame natural bottlenecks in the system to keep things scalable?

When something in the framework doesn’t make much sense from an architectural or coding perspective, it usually means there should be a better way. One example comes to mind: we wanted to process thousands of requests and we were hitting our data provider too hard, which led to throttling. But most of these requests were actually exactly the same, due to the way we parallelize the work. So we came up with a short-lived data caching layer that rendered the problem almost non-existent.

Q: Insurance and risk management depend on accurate data. How does Arbol's approach to data management stand out?

We rely on a team of professional data engineers, whose sole responsibility is to create smooth and flexible processing pipelines of publicly available datasets. First, this helps us to be confident that data itself will be up to a high standard and will be accepted by the community and, second, that the data will fit our needs.

Q: Please tell us about your ongoing interest in Diffusion and Consistency models. For the benefit of our readers, can you unpack these concepts and explain why they're essential to today's risk management landscape?

These two types of models come from the same paradigm, which is known as Generative AI. They let everyone create new visual content, based on certain text prompts. We believe that Generative AI is a very promising technology that will let us explore the climate risk landscape in a more robust way by simulating certain conditions, primed with our meteorological knowledge.

Q: Given the rapid evolution of technology and its intersection with risk management, where do you see the industry's future going?

I think the industry will gravitate towards more AI systems, and that professionals will help guide these systems. We’ve already seen a demand for new professions like prompt engineers. In our domain, one could imagine a meteorologist using AI-driven applications to assess different scenarios for how climate change may evolve.

Q: You recently attended the International Conference on Machine Learning. Were there any key takeaways or insights from the conference that you hope to implement at Arbol in the near future?

There were several papers presented there that I found interesting, which closely relate to what we do. One of the papers dwelled on the idea of blending many different inputs via an advanced Deep Learning model, which may turn out to be a promising approach.

Q: Lastly, for young professionals looking to delve into the intersection of tech, data, and risk management, what advice would you give them?

I’d recommend learning more about AI and the most recent developments coming out of this field, which will enrich the toolbox of any young professional.

To stay in the loop with all of the exciting initiatives underway at Arbol, we invite you to follow us on LinkedIn and X.