Summary
Satyam was the first of its kind system for generating groundtruth annotations for ML vision tasks (pre-sagemaker). The key advantage of Satyam is that it ironically while it employs untrained mechanical turk workers for vision tasks, it completely automates the rest of the chores -- image and video pre-preocessing, HIT generation and launching, quality control, payments, pricing and worker filtering.
In other words you simply fill out a web form, indicating the location of images in your Azure blob store, your Amazon mechanical turk account and customize instructions for your task and hit launch! Satyam takes care of the rest and deposits the annotated groudtruth (in JSON) at the azure blobstore you provided.
It was used at MSR internally for ML video analytics research initially, but now has been productized and some of its features such as automated quality control are part of the Azure Labeling Service within the Azure ML pipeline.
Watch the video behind Satyam
