In this paper, we present CACHE-ED, a novel framework for document entity extraction that combines the power of large language models (LLMs) with graph-based document representations, caching mechanisms, and an actor-critic multi-agent architecture. Our approach addresses the inefficiencies and inaccuracies that are common in extracting structured information from documents, particularly in templated formats like invoices. CACHE-ED implements a human-in-the-loop paradigm in which human reviewers validate and update the system’s outputs, establishing a feedback loop that progressively enhances the accuracy of future extractions. This process ultimately eliminates the need for human intervention over time and optimizes tail-end accuracies, refining the system from 90% to near-perfect precision. Our experiments demonstrate that this approach outperforms industry-wide used extraction mechanisms by 8% and improves the speed & reduces cost by over 50% each, making it a scalable solution for real-world applications.