Thank you! Tbh I really over-engineered it at first, trained a model to classify articles into sub-categories, then built a query around that to fetch contextualized videos from pexels, but it was really clunky (e.g., stock market could return something like someone buying fruit at the farmer's market). Currently, I have a 2 pools of videos I curate manually, and the content creator script just picks randomly either from the stocks pool, or crypto pool.