OGRRE: Enabling A Human-in-the-Loop for AI/ML Driven Oil & Gas Data Extraction UI

This poster describes the research software engineering and user experience approaches, as well as lessons learned from initial deployments, for a new open-source web UI tool, OGRRE, which enables rapid human review of Google Doc AI-digitized oil and gas regulatory documents.

The US has hundreds of thousands of orphaned oil and gas wells that no longer have a responsible owner. These wells pose an environmental and human health risk, as they can leak methane and other pollutants into the air and groundwater if they become compromised. In 2021, the Bipartisan Infrastructure Law was passed which allocated $4.7 billion for orphaned well plugging in the US. One major barrier to the Herculean task of plugging these wells is providing the stakeholders doing this work with accurate information about their location, construction, and status. This information is often documented in regulatory paper records created during the well permitting process. While many regulatory agencies have scanned their paper records, they often lack the resources necessary to digitize their content. Extracting this information and organizing it in computer databases will help regulatory agencies as they seek to better characterize the orphaned wells under their jurisdiction and prioritize their plugging.

Modern AI/ML optical character recognition (OCR) models can aid the efficient digitization of historic well records. However, these documents often date back over 100 years and contain handwritten fields and many other anomalies. This results in a significant challenge when training AI/ML algorithms to accurately identify fields, and results are often imperfect. Human review of the documents is needed to find and correct problems. With hundreds of thousands of documents to process and limited resources, an efficient melding of the human-in-the-loop tasks with the AI/ML models becomes essential.

This poster will describe the Oil and Gas Regulatory Record digitizEr (OGRRE) — a custom user interface we developed to facilitate rapid human review of digitized oil and gas regulatory documents. OGRRE uses retrained AI/ML models developed by Google Document AI. The tool is designed to  enable the interactive review and correction of AI/ML extracted data. The poster will describe the combination of Google Doc AI modeling, software engineering, and user experience approaches used to create the OGGRE tool and software infrastructure, as well as lessons learned from our pilot deployment of the tool.

OGRRE development is supported by the United States Department of Energy’s Undocumented Orphaned Well Program under the CATALOG[1] project.