Package com.tnsai.evaluation.evaluators.safety

package com.tnsai.evaluation.evaluators.safety

Safety evaluation metrics for LLM outputs.

Evaluators in this package detect hallucination, bias, and toxicity in agent responses using LLM-as-judge methodology.

Since:

0.3.0

Related Packages

Package	Description
com.tnsai.evaluation.evaluators	Layer 1: Evaluators that score agent behavior on specific dimensions.
com.tnsai.evaluation.evaluators.agent	Agent-tier evaluators (TNS-373).
com.tnsai.evaluation.evaluators.geval	G-Eval: LLM-as-judge evaluation using chain-of-thought scoring.
com.tnsai.evaluation.evaluators.multiturn	Multi-turn conversation evaluation metrics.
com.tnsai.evaluation.evaluators.rag	RAG (Retrieval-Augmented Generation) evaluation metrics.

Class	Description
BiasEvaluator	Detects demographic, cultural, or ideological bias in agent responses.
HallucinationEvaluator	Detects hallucinated content in agent responses by checking factual claims against provided context.
PIILeakageEvaluator	Detects PII leakage in agent responses using regex + checksum validation.
PIILeakageEvaluator.Category
ToxicityEvaluator	Detects toxic, harmful, or offensive content in agent responses.