Energy market information providers in the US had a specific requirement of extracting and processing information from complex documents, including tables, notes, and subsidiary data, is often labor-intensive and prone to errors. Current systems struggle to efficiently activate OCR layers for images and apply relevant business rules and formulae. We propose a model trained with Stochastic Gradient Descent (SGD) to automate these tasks, featuring capabilities for information extraction, rollback, and relearning to enhance accuracy and adaptability, seamlessly integrating into existing workflows.