General Finance
Mar 20, 2025
Get to Know Gradient: Leo Pekelis, Chief Scientist
Where were you before Gradient?
“Before leading research and analytics at Gradient, I held data leadership positions with CloudTrucks, Disney, and Optimizely. I earned a PhD in statistics from Stanford and have been doing my best to apply data-centric products wherever they might be most valuable. This ultimately led me to artificial intelligence and Gradient.”
What inspired you to join Gradient’s team?
“Gradient is an incredibly exciting place to be right now because it’s solving a unique problem with the right tools at the right time. That’s a rare combination.”
“Gradient’s target industry – finance – is very well-suited to benefit from the bleeding edge of AI technology. The finance industry is intelligence-powered - it relies on a ton of information, deep research, and complex reasoning to make decisions. But, this process is bottlenecked by messy, unstructured data that requires considerable manual effort to comb through.”
What excites you about the intersection of AI and finance?
“The potential for AI to revolutionize the way that finance professionals work.”
“The scope of work when making investment decisions almost always requires searching deeply through long documents like 10k reports or earnings transcripts and synthesizing results. This is exactly the problem that large language models are best suited to solve – particularly the ones that Gradient has developed - and has large efficiency gains in saving an analyst’s time.”
What’s some advice you have for companies looking to implement AI?
“Don’t underestimate the importance of the discovery process. Information sharing between your company and ours helps ensure that everyone is aligned in terms of needs, goals, success metrics, and capabilities.”
“In general, there tends to be a disconnect between the excitement around AI and the familiarity of how to actually use it in a repeatable and well-integrated way. Spending time sharing information during the discovery process can help both companies reduce this disconnect and speed up the time to market.”
“For example, data cleanliness matters a lot when implementing AI. So, if a company is looking to automate a frustrating process – like an intelligence workflow – it’s best if they have clear examples of inputs and consistent outputs. This makes it much easier during the training process to solve the task at hand.”
What exactly do you mean by “clean data”?
“A definition that I’ve found useful is data that is relevant, recent, concise, and consistent.”
“Gradient’s AI can interpret text, pictures, and numbers so the physical format doesn’t matter as much. But, the relevancy of that data is very important. Without relevant data, the AI will have a harder time learning the nuance of what is asked from it, and so the outputs will be less accurate. Similarly, recency, conciseness, and consistency help to reinforce positive patterns without distracting the models with irrelevant information.”
What’s one of the most recent trends that you’ve seen in regards to AI?
“Access to AI is getting democratized very quickly.”
“Just a few years ago, implementing AI required extensive domain expertise and engineering capabilities. This meant that the technologies could only be used in very specific areas, like sentiment analysis as part of a larger data processing pipeline with engineered features fed into a machine learning model. But, most AI tools are now capable of processing natural human language, making them much easier to use and bringing the power of AI to a wider audience.”
“We are quickly reaching a point where not only the access to sophisticated AI is democratized, but their capabilities as domain experts and software engineers are strong enough that through them, building and implementing AI technologies is itself democratized.”
What skill is becoming more important due to AI?
Clear communication.”
“Being able to clearly tell a large language model what you need will be an immensely valuable skill moving forward. Providing crystal clear instructions to the model is usually the difference between getting the answer you want and getting the wrong answer. The structure of prompts and verbiage of interrelated sections can be especially important. LLMs are eager to extract patterns, and often don’t have the same context as the writer to tell which ones are unintended.”
What are some of the biggest hurdles companies face when adopting AI?
“One big hurdle is how quickly AI technology is evolving.”
“The rapid rate of change can make it difficult for organizations to adopt AI because its capabilities are constantly changing. This is why it can be beneficial to partner with an AI provider, as the provider updates and expands how it leverages AI technology over time. What we’ve heard from customers is this is a big value-add that Gradient provides.”
Thanks for reading!
We hope that you’ve enjoyed this episode of Get To Know Gradient as we interviewed Gradient’s Chief Scientist, Leo Pekelis.
Share
