Clarify and define Llama Stack personas

[rhuss]

Afais Llama Stack targets multiple personas, but please correct me when I'm wrong:

AI Engineer / Developer / AI-enabled Business Developer / ...

• User of the LLS API endpoints for inference, agentic, evals
• Integrates calls to LLS API into their applications
• Uses the chat UI of LLS
• Authenticates against an IDP to access the LLS API endpoints

Data Scientist / ML Engineer / ..

• Users of the LLS API for post-training and SDG
• (same uses cases as for an AI engineer)

MLOps

• Responsible for operating an LLS server
• Selects and wires together the available backend providers
• Creates and maintains LLS distributions (i.e. the persona that calls llama stack build and llama stack run)
• Responsible for setting up the AI infrastructure, like the backing databases, references to the inference server, etc.
• Use the infrastructure to install stuff (e.g., has access to the K8s API), and authenticates against the infrastructure itself.

(I'm ok to combine the Data Science and AI-Engineer personas if needed or if the separation for the LLS API makes no sense, but the MLDevOps and AI-Engineer personas should be separated. Note that a person can have multiple personas, e.g. when running locally on a developer notebook, your are typically both, the user (AI engineer) and admin (MLOps)

I think, for every feature that we include, we should explicitly mention the persona that this feature addresses. This avoids some confusion and misconceptions.

For example, currently, I'm struggling with which persona is responsible for registering tools. Is it the AI Engineer via the LLS API? Or the MLOp who configures all allowed functions up-front? Or is it both? (but then there should be support for both use cases, API-based function and configuration-based function registration)

wdyt, would it make sense to shape out and surfacing personas more prominently?