TruLens Eval

TruLens-Eval

Known Issues

• Awaitable and generator method inputs/outputs not properly recorded.

Work Streams

Features

• Conversation/session tracking, display, feedbacks.

Feedback analysis interfaces

• Correlations between feedback results.

• Feedback clustering (f1, f2 behave similarly, f3, f4 behave similarly).

• Derived metrics/feedbacks: e.g. "Does high context relevance imply low abstain?"

• Statistical validity computation and presentation.

Additional feedback functions

• Safety with LlamaGuard

OpenTelemetry interfaces

• Goal: Export records to otel.

• Goal: Record via otel interface.

• Goal: Record and view other tool's instrumented spans alongside/within ours.