For me it’s posterior sampling to generate meaningful uncertainty metrics in very large scale causal inference for understanding location-based behavioral data, sometimes for ads, sometimes for other modeling problems like image quality, app engagement, etc.
