Computing recommendations on large datasets can be very slow. To account for this, we can have Lux render tabs when they are computed instead of rendering Lux only after all tabs are computed; that is, lazily stream in the tabs/recommendations. This will allow users to interact with a part of the widget while the widget computes the rest of the recommendations.