Pipeline explainability

[fvanlitsenburg]

Situation
For my website (www.wetzoek.nl) I use a Haystack neural search engine that's relatively complex.

Problem
The results it turns out are not always what they should be. This could have many causes:

1. Wrong choice of model / model type
2. Mistake in the pipeline
3. Mistake in document embedding
4. Model not finetuned (enough)
5. Poor pipeline setup

Solution
Functionality to explain the outputs Haystack gives me.

[ZanSara]

Hey @fvanlitsenburg, interesting idea! How do you imagine such feature to look like? A sort of diagnostic tool for pipeline configurations? Or you are thinking more about pipeline outputs?