Abstract
Transactional Systems are ERP systems that offer various services for the Oil and Gas industry, such as procurements, warehousing, plant maintenance, supply chain, finance, sales, and marketing. However, interacting with transactional systems screens can be challenging for users who are not familiar with the technical details. In this paper, we propose TransLing, a language model architecture that train, fine-tune and deploy smaller LLM models to enable natural language interaction with transactional screens in the Oil and Gas industry such as SAP Screens. TransLing can generate values for user input screens based on the user questions in natural language, translate natural language questions to possible ad-hoc reports, and find transactions from the description. TransLing outperforms the baselines in terms of accuracy and usability and simplifies the technical details and the specific terminology of transactional screens for users who are not familiar with them.
