Package com.tnsai.coordination.judge

Class LLMJudgePolicy

java.lang.Object

com.tnsai.coordination.judge.LLMJudgePolicy

All Implemented Interfaces:

JudgePolicy

public final class LLMJudgePolicy
extends Object
implements JudgePolicy

A JudgePolicy that uses an LLM to evaluate coordination output.

The LLM receives the output along with configurable evaluation criteria and produces a structured verdict. This enables natural-language quality assessment for complex, subjective criteria.

Usage

 LLMClient judge = new OpenAIClient("gpt-4o", apiKey);
 JudgePolicy policy = LLMJudgePolicy.builder()
     .llmClient(judge)
     .criteria("Evaluate for correctness, completeness, and clarity. " +
               "Score 0.8+ means production-ready.")
     .build();
 

Since:

0.2.3

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static final class	LLMJudgePolicy.Builder	Builder for LLMJudgePolicy.

Method Summary

Modifier and Type	Method	Description
static LLMJudgePolicy.Builder	builder()
JudgeVerdict	evaluate(String agentId, Object output, JudgeContext context)	Evaluates an agent's output and produces a verdict.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Details

evaluate

public JudgeVerdict evaluate(String agentId,
 Object output,
 JudgeContext context)

Description copied from interface: JudgePolicy

Evaluates an agent's output and produces a verdict.

Specified by:

evaluate in interface JudgePolicy

Parameters:

agentId - the identifier of the agent whose output is being judged

output - the output to evaluate (type depends on coordination pattern)

context - iteration state and history

Returns:

a verdict with decision, score, and optional feedback

builder

public static LLMJudgePolicy.Builder builder()