GOAL-BASED AGENTIC AI FOR AUTOMATING CLAIM VALIDATION USING LARGE LANGUAGE MODELS

Abstract

This paper presents a goal-based agentic AI system for automating individual forest claim validation under India’s Forest Rights Act (FRA). The system integrates legal rule enforcement through formal logic gates with interpretive reasoning by a large language model (LLM). Each claim is evaluated across statutory conditions such as identity, occupancy date, land extent, and evidence sufficiency. A structured prompting and tool-augmented workflow enables the LLM to reason through unstructured data while maintaining compliance with FRA criteria. In simulation using 1,000 synthetic claims, the proposed agent achieved 92.4% accuracy—outperforming baseline rule systems and zero-shot LLM classifiers. The agent's decisions are explainable, auditable, and adaptable, demonstrating how hybrid AI architectures can enhance legal workflows. The model’s modularity supports future extensions to community claims and other entitlement domains. This work contributes to the emerging field of AI-assisted legal adjudication, particularly in public governance contexts involving high-stakes land rights.