Self-Evaluating Agentic Framework Using Retrieval-Augmented Generation for Accountable Environmental Sustainability Communication

Mukesh Patel School of Technology and Management, NMIMS, Computer Engineering Department, 400056 Mumbai, India

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Published online: 29 May 2026

Abstract

Effectively communicating information about environmental sustainability is challenging because of complicated policies and multiple stakeholder inquiries. In this paper, we propose a unique agentic framework that brings together meta-classification, retrieval-augmented generation (RAG), and self-evaluating large language models to accomplish that. During the pre-call stage, user inquiries about environmental policy, or what the organization is doing sustainability-wise, are processed through a meta-classifier to either a RAG system, drawing on a knowledge base, or a standard large language model (LLM), with the result surfaced in a dashboard for agents to use. The audio of the calls is then transcribed using Whisper in the post-call stage, and the output processed in a generative evaluation loop where a generator LLM assesses the responses, while another LLM acts as an AI-as-judge to provide feedback about performance. Experimental outcomes have indicated enhanced accuracy, relevance, and trustworthiness, indicating the framework's success in enabling accountable high quality communication of environmental sustainability information.