Leveraging State-of-the-Art Generative AI Models for Operational Excellence: Summarizing Daily Operational Remarks

The PETRONAS leadership team requires a swift and precise information flow about daily operations for decision-making but often, operational remarks from engineers are too verbose. Therefore, a smart summarizer using generative AI techniques was proposed, which refines and optimizes the remarks to present only the critical and relevant information to the leadership. The main pain points identified and addressed in this work include the lengthy and repetitive nature of text information, and the necessity for accurate entity recognition and contextual comprehension. The summarizer uses the OpenAI’s GPT-4 0613 model for abstractive summarization, semantic search, and output formatting in solving the said pain points. Various prompt engineering techniques were evaluated and refined, and the final model blends zero-shot prompting, few-shot prompting, and prompt chaining. Additional model parameters were introduced and tuned to achieve output relevancy and consistency, and data processing techniques were used to improve runtime. Manual result validations were conducted with subject matter experts to ensure accuracy, relevance, readability, and output format, which resulted in an 82% acceptance rate.