Five people, from firmware to the API your users hit, all named below. You talk to the people doing the work.
Your point of contact from the first call to handoff. Scopes the audit, runs the project, and keeps the numbers honest. Knows GPUs and inference well enough that scoping is a technical conversation.
Runs the engineering team and writes most of the code, from CUDA kernels and low-level code up through infrastructure to the full stack.
Code that runs on bare metal: firmware, embedded systems and the boards themselves. Takes a model off the cloud and onto the device when inference has to run at the edge.
The services around the model: the APIs, queues and storage that put an optimised model in front of real users, and keep it there.
Reviews every access grant and every change we make inside your environment, and makes sure our access is gone at handoff.