Locate the application run
Start with the flagged answer and open the model trace associated with that interaction.
Sense / AI operations
Inspect the model interaction behind the answer.
Sense brings model traces and evaluation context into Unified Observability. Engineers can examine how an AI application assembled an answer and connect a quality concern to the model interaction and application behavior that produced it.
AI & LLM observabilitySense
Inspect the support-summary run flagged for an unsupported refund statement.
In the workspace
Application run
support-summary · run-summary-d62
Model context
Retrieved policy and generated answer
Evaluation review
Refund statement · Source support
Examine the answer alongside the context supplied to the model.
How it works in Sense
The engineer can assess whether the refund statement is supported by the supplied policy and identify which part of the application needs investigation. The evaluation flag remains a review signal until the underlying interaction has been inspected.
Inspect the supplied context
Review the retrieved policy and model output to understand what evidence was available when the answer was generated.
Assess the quality concern
Compare the disputed statement with its source support and prepare a focused change to retrieval or answer generation for testing.
Bring Sense to your team.
Explore ai & llm observability around your company’s workflows.