Natural language processing and machine learning models for non-compliance detection

Abstract

Systems and methods are disclosed for performing automated compliance classification of natural language text using a Large Language Model (LLM)-based reasoning–action– observation framework that instructs an agentic framework. The system may access a review item comprising text and associated metadata, generate one or more prompts defining a computational sequence of reasoning, action, and observation operations, and execute a reasoning process to form hypotheses identifying potential compliance violations and missing data. The system may construct a tool call specifying a target agent within the agentic framework to perform retrieval, extraction, alignment, or classification operations, receive structured responses including factual or confidence data from a data warehouse, and integrate those responses into updated prompts for continued reasoning cycles. The system may iteratively refine results until a stopping condition is met, producing a compliance classification record that includes predicted findings, policy identifiers, supporting evidence, and calibrated confidence values.

Images