Runtime Evaluation¶
Evaluations play a crucial role in improving LLM app outputs by altering execution flow at runtime.
TruLens supports runtime evaluation via two different mechanisms:
- In-line Evaluations - evaluations that are executed during an agent's execution flow and passed back to agent to assist in orchestration.
- Guardrails - evaluations that can be used to block input, output and intermediate results produced by an application such as a RAG or agent.