api-server needs to provide one or more endpoints that enable a user, via the HTTP api, to specify a datafile, load it with GPU I/O into a column, and then serialize the column over arrow to a caller. I'd prefer to use a generic API that lets any @rapidsai/cudf API call to be used.

• Find a good dataset for the particles demo.
• Create an api-server endpoint that loads the file using one or more @rapidsai/cudf commands, all visible on the HTTP api.
• Create an api-server endpoint that lets the user reflect on data that has been loaded onto the GPU process.
• Create an api-server endpoint that streams the loaded data specified by the user to a client via arrow.