feat: Add cookbook for Hallucination Risk Calculator

[sjrl]

Cookbook for this PR deepset-ai/haystack-experimental#359

Some feedback on this from @julian-risch deepset-ai/haystack-experimental#359 (comment)

Would be great to have also a RAG Pipeline example that uses a DocumentStore and a Retriever. It doesn't stop us from merging this PR but I believe it would help users to try out the new feature and we should add it in another PR.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[sjrl]

@julian-risch this is ready for review! It includes an example now where we create a RAG Query pipeline

[julian-risch]

LGTM! Except for my comment above, I don't have any comments on how to improve. The RAG example is really great to have! The output cell at the very end with an example of no answer due to missing relevant documents too.
Disclaimer: I did not run the code.

[bilgeyucel]

Why is there a risk here although the answer is basically the evidence provided in the prompt?

Reply via ReviewNB

[sjrl]

It's a risk bound and the risk is assessed based on this number and a threshold. I can't say I fully understand how it works this was mostly copied from the original repo.

[bilgeyucel]

RAG-based Example

Reply via ReviewNB

[bilgeyucel]

Left my comments! Can you also add the notebook to the index.toml for it to be rendered on the website. Here's an example for notebooks with experimental components.

[julian-risch]

Experimental release is done. I'll merge this notebook so that we can link to it in release notes.