I became a data scientist because I liked getting my hands dirty with data, building models, and slowly turning a mess into something understandable. Then LLMs arrived and seemed to reduce all of that to typing a prompt. I was not impressed. What changed my mind was not chatbots, agents, or generated code. It was structured output. LLMs can turn sources that were previously unusable at scale into datasets we can validate, query, visualize, and learn from. In this talk, I will move from a tiny extraction example to analyzing hundreds of podcast episodes, then mapping the characters, travel, technology, and plot of my favorite sci-fi books, and finally show how the same pattern applies to real security work at Neo Security. This is how LLMs made data science fun for me again, not by giving me answers, but by helping me create entirely new data.
I am not a bot, hopefully.