Realistically what's the biggest you can make the dataset if you need to prebuild everything? Aggregation and filtering help, but quickly become impractical if you want people to change the filters dynamically.