Currently we are using the following repo for pose estimation.
The code sits in a container inside an Amazon Ubuntu intance. (We have to set up a good way with which we may all access this intance.)
Currently we can run the binary on videos and a get a Json string with the estimated poses. Soon functionality will be fully available throught (most likely) a python api.