Ideas: - Move processing to bare metal cluster (more resources) - Look into node's worker threads - Look into gRPC (stream the progress to the client?)
