Improve the UX the comment section for LLMaJ evaluators.

[dahnny012]

Describe the feature or potential improvement

Self Hosted - Langfuse 3.131

Problem

Hello currently we run experiments on langfuse datasets via a python service that has code to:

1. Call our LLM application
2. Get the Trace for it
3. Run our custom evaluators
4. Create a dataset run
5. Upload the scores

This has been an ok experience however there have been some UX challenges. For example its a bit difficult to peek at the reasoning for a score.

You have to hover over it and the box cuts off. It be ideal if this was presented as a modal of some sort. As an alternative it seems like you can go to the trace and see it in the scores panel but its an extra click for information that is already there.

More Questions

We implemented this evaluation loop before the experiments api was released. Would the observability of evaluations improve if we migrated to the new experiments api?

Additional information

No response

[marliessophie]

Thanks for sharing feedback here! We have the goal to invest more and more into UX over the next weeks and months.

Have you tried clicking the compare button in the top right corner? This should yield a slightly improved view. The score details (eg reasoning) still only appear in a hover card though but should at least not be cut off. Any feedback on that experience too would be very helpful, thanks!

[dahnny012]

That is slightly better but in the comparison view I lose the filter I had running. The use case here was to investigate failed dataset item runs and moving to the compare view removes that filter.

Also waiting for the hover is a bit of a bad UX. The reasoning tab could be in its own column in the comparison.

[marliessophie]

Appreciate the feedback. What is the filter you had set? You can filter by scores for a given run, might this be helpful?