Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AEMA: Verifiable Evaluation Framework for Trustworthy and Controlled Agentic LLM Systems

YenTing Lee, Keerthi Koneru, Zahra Moslemi, Sheethal Kumar, Ramesh Radhakrishnan

TL;DR

Evaluating agentic LLM systems in multi-agent workflows requires verifiable, auditable processes rather than single-turn scoring. The authors present AEMA, a process-aware framework with Planning, Prompt Refinement, Evaluation, and Final Report agents that plan, parameterize, score, and summarize multi-step agent actions, all under human oversight. In enterprise-style finance workflows, AEMA demonstrates higher stability and closer alignment to human judgments than a single LLM-as-a-Judge, while providing auditable evaluation traces and adaptable domain planning. The work argues for cross-domain applicability and outlines future directions to improve efficiency, cost, and generalization, enabling more trustworthy and controllable autonomous AI systems.

Abstract

Evaluating large language model (LLM)-based multi-agent systems remains a critical challenge, as these systems must exhibit reliable coordination, transparent decision-making, and verifiable performance across evolving tasks. Existing evaluation approaches often limit themselves to single-response scoring or narrow benchmarks, which lack stability, extensibility, and automation when deployed in enterprise settings at multi-agent scale. We present AEMA (Adaptive Evaluation Multi-Agent), a process-aware and auditable framework that plans, executes, and aggregates multi-step evaluations across heterogeneous agentic workflows under human oversight. Compared to a single LLM-as-a-Judge, AEMA achieves greater stability, human alignment, and traceable records that support accountable automation. Our results on enterprise-style agent workflows simulated using realistic business scenarios demonstrate that AEMA provides a transparent and reproducible pathway toward responsible evaluation of LLM-based multi-agent systems. Keywords Agentic AI, Multi-Agent Systems, Trustworthy AI, Verifiable Evaluation, Human Oversight

AEMA: Verifiable Evaluation Framework for Trustworthy and Controlled Agentic LLM Systems

TL;DR

Evaluating agentic LLM systems in multi-agent workflows requires verifiable, auditable processes rather than single-turn scoring. The authors present AEMA, a process-aware framework with Planning, Prompt Refinement, Evaluation, and Final Report agents that plan, parameterize, score, and summarize multi-step agent actions, all under human oversight. In enterprise-style finance workflows, AEMA demonstrates higher stability and closer alignment to human judgments than a single LLM-as-a-Judge, while providing auditable evaluation traces and adaptable domain planning. The work argues for cross-domain applicability and outlines future directions to improve efficiency, cost, and generalization, enabling more trustworthy and controllable autonomous AI systems.

Abstract

Evaluating large language model (LLM)-based multi-agent systems remains a critical challenge, as these systems must exhibit reliable coordination, transparent decision-making, and verifiable performance across evolving tasks. Existing evaluation approaches often limit themselves to single-response scoring or narrow benchmarks, which lack stability, extensibility, and automation when deployed in enterprise settings at multi-agent scale. We present AEMA (Adaptive Evaluation Multi-Agent), a process-aware and auditable framework that plans, executes, and aggregates multi-step evaluations across heterogeneous agentic workflows under human oversight. Compared to a single LLM-as-a-Judge, AEMA achieves greater stability, human alignment, and traceable records that support accountable automation. Our results on enterprise-style agent workflows simulated using realistic business scenarios demonstrate that AEMA provides a transparent and reproducible pathway toward responsible evaluation of LLM-based multi-agent systems. Keywords Agentic AI, Multi-Agent Systems, Trustworthy AI, Verifiable Evaluation, Human Oversight
Paper Structure (19 sections, 1 equation, 2 figures, 2 tables)

This paper contains 19 sections, 1 equation, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work: From Single-Turn Evaluation to Process-Level Assessment of LLM Multi-Agent Systems
  3. Method
  4. AEMA Framework
  5. Planning Agent
  6. Prompt Refinement Agent
  7. Evaluation Agents
  8. Final Report Agent
  9. Experiments
  10. Experiment Settings
  11. Formal Plan-Evaluation Metric
  12. Results and Analysis
  13. Planning Experiment.
  14. Stability Experiment.
  15. Human Alignments Experiment.
  16. ...and 4 more sections

Figures (2)

  • Figure 1: Overview of AEMA (Adaptive Evaluation Multi-Agent): Planning Agent builds the plan; Prompt-Refinement Agent retrieves and prepares examples; Evaluation Agents score intermediate actions; Final Report Agent aggregates results into an auditable, reproducible report.
  • Figure 2: Stability of final scores across 30 evaluations. AEMA exhibits lower dispersion than a single LLM-as-a-Judge.