General Report Automation for Problem and Hazard Observation in Systems (GRAPHOS)

Machina Cognita Technologies (MCT) proposes General Report Automation for Problem and Hazard Observations in Systems (GRAPHOS), a Large Language Model (LLM)-based solution for the automated generation of enhanced, standardized problem reports for Aegis Combat System (ACS) in navy surface platforms. GRAPHOS will incorporate an array of task-specific LLMs. First, a Document-Standards Engine will interpret and enforce document structures based on existing, high-quality report examples. Secondly, a Report-Synthesis Engine will integrate analytic insights and contextual metadata into generated reports that adhere to identified document standards regarding content and format. And finally, as an option-period extension, a Machine-Learning (ML)-Configuration Engine will build pipelines for training models to detect anomalies in live ACS data streams automatically. Upon operator-triggered or automatic anomaly detection, GRAPHOS will solicit problem summaries from operators, identify correlated cases from a centralized database describing prior anomalies, and apply the generated document standards for problem report structure. GRAPHOS will then output and store standardized reports to facilitate holistic reports analysis. GRAPHOS will ultimately ease the reporting burden for personnel and accelerate ACS testing and development, thereby improving system performance in navy surface platforms. The technical work of GRAPHOS will develop new key capabilities. First, GRAPHOS will offer automated creation and enforcement of document standards via the Document-Standards Engine. This engine will use a finetuned Document-Standards LLM to understand operator objectives, select relevant examples from previous problem reports, and generate explainable report-generation instructions and template rules. This approach fully automates the creation and application of document standards for various types of reports. Secondly, the GRAPHOS Report-Synthesis Engine will integrate consolidated sensor data and metadata into LLM-generated reports, ensuring adherence to document standards and consistency in measurements. This agile problem-reporting capability enhances explainability by incorporating contextual information from data analysis results. Thirdly, GRAPHOS will offer holistic analysis of underlying trends and patterns across levels and sites. And finally, the GRAPHOS ML-Configuration Engine will develop an LLM for configuring and executing ML training pipelines to build anomaly-detection models for ACS data streams. By automating the selection of model types, feature extraction methods, hyperparameters, and other critical choices, this approach significantly reduces manual effort while providing explanations for each generated design specification.