Yes, human reviewers can directly add evaluations in Phoenix. Phoenix supports human annotations through its UI, allowing users to add, edit, view, and delete annotations on traces and spans. This can be useful for manual quality assurance and sharing insights within a team. According to <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/llm-traces/how-to-annotate-traces.md|Annotating Traces>, Phoenix provides a simple mechanism to collect user feedback, such as thumbs-up or thumbs-down buttons, which can be logged as annotations.
Alternatively, you can collect evaluations through another tool and import the results into Phoenix. This can be done by creating a DataFrame with the necessary evaluation data and using the `log_evaluations` method to upload it to Phoenix. This approach is useful if you have a large volume of evaluations or if you prefer using external tools for data collection. For more details, see <https://github.com/Arize-ai/phoenix/blob/main/docs/evaluation/how-to-evals/evaluating-phoenix-traces.md|Evaluating Phoenix Traces>.
Would you like more detail on either approach?
Sources:
- <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/llm-traces/how-to-annotate-traces.md|Annotating Traces>
- <https://github.com/Arize-ai/phoenix/blob/main/docs/evaluation/how-to-evals/evaluating-phoenix-traces.md|Evaluating Phoenix Traces>
