Ferdi Kossmann (Sovara Labs)
Many organizations have accumulated data for decades, organically evolving what they collect, how they organize it, and which systems they use to process it. This results in data silos, inconsistent naming, stale and duplicated versions of the same data, and a landscape that is hard to navigate in general. To address this, companies undertake extensive efforts to clean up their databases and make them manageable again. In some cases, this costs millions of dollars and takes years to complete, while the underlying schemas, systems and business logic continue to change.
Yet new employees learn to navigate this same data within weeks or months, without first reorganizing it. In this talk, I explore whether AI can do the same: gradually learn where things are, what they mean, and how they are used, working with the data directly as it already exists. I will discuss the technical challenges associated with this and the approach we take at Sovara Labs to solve them. I will then give a demo and show examples of how Sovara is used in real-world applications.
Ferdi Kossmann recently completed his PhD at MIT, where he worked closely with AI systems in real-world settings. Through this experience, he saw how even sophisticated AI agents struggle with messy organizational data that requires context, conventions, and tribal knowledge to navigate, and how difficult it is for teams to clean and organize that data for AI. He is now co-founder and CEO of Sovara Labs, a stealth startup building AI systems that learn to work with a company’s data as it already exists.