Documentation: Add experiment creation API call to evaluation guide

[pierrevalade]

Issue Description

The evaluation guide at shows how to run evaluations with experiments, but doesn't demonstrate how to actually create an experiment via the API.

Current State

The documentation currently:

• Shows generating an experiment ID locally: experiment_id = f"email-eval-{datetime.now().strftime('%Y%m%d-%H%M%S')}"
• References using this ID in completion metadata
• Shows sending annotations for the experiment

However, it's missing the actual API call to create the experiment entity.

Proposed Improvement

The evaluation pipeline script should:

1. Create a new experiment via the experiments API
2. Add completions to this experiment (already shown)
3. Send annotations for evaluation results (already shown)

This would provide a complete end-to-end example of programmatically running experiments.

Affected Section

Lines 471-472 in the "Create the Evaluation Pipeline" section currently show only local ID generation without the API call.

-- Claude Code

[pierrevalade]

@anyacherniss sorry I missed that part during the review of this new documentation PR, for this page:

https://docs.anotherai.dev/use-cases/fundamentals/evaluating

based only on the code shared on the current page, I don't think you can get to a experiment page, (as shown in the last screenshot though, so why I got confused), because the code to create a new experiment via the API* is never included in the code sample.

* this API endpoint https://api.anotherai.dev/docs#/default/create_experiment_v1_experiments_post

@anyacherniss let us know if you need some clarification.

cc @guillaq

[anyacherniss]

I can definitely ask Claude to help me get this content added.

I am a little confused about the following (+ would like to understand so I have a better understanding of how this aspect of AnotherAI works):

based only on the code shared on the current page, I don't think you can get to a experiment page, (as shown in the last screenshot though, so why I got confused), because the code to create a new experiment via the API* is never included in the code sample.

To confirm: you mean the experiment with the scores is only possible to be created via MCP? If yes, I am confused by this statement because I asked Claude Opus (used via Cursor) how it created an experiment with scores (not the exact one shown in documentation, but one I created yesterday on staging) and Claude said the following:

So to directly answer: I created the experiment structure via MCP tools, but had to use the REST API endpoint directly for adding the accuracy score annotations because of parameter formatting issues with the MCP annotation tool. The combination of documentation, MCP tools for the main flow, and direct API calls for annotations got the job done!

So it sounds like it was able to create the experiment via MCP, and only needed the API for the annotations? But maybe it's approach was hacky/incorrect?

[pierrevalade]

So to directly answer: I created the experiment structure via MCP tools, but had to use the REST API endpoint directly for adding the accuracy score annotations because of parameter formatting issues with the MCP annotation tool. The combination of documentation, MCP tools for the main flow, and direct API calls for annotations got the job done!

So it sounds like it was able to create the experiment via MCP, and only needed the API for the annotations? But maybe it's approach was hacky/incorrect?

OK, yeah, I understand better now. I never thought about this hybrid approach where the MCP client is creating the experiment via a tool call, but the (Python) is running the completions... the fact that this approach worked for you @anyacherniss, validates by definition the approach... look like there are many ways to get to the same outcome. I'll need to think more...