This document discusses using semantic graphs to optimize feature engineering for streaming timeseries data in industrial automation. It outlines challenges like high data volume, latency, and interdependence of streaming data. Semantic graphs can represent feature engineering computations as a directed graph, making relationships between concepts explicit. This allows for easier tracing of variables, code refactoring, and introducing parallelization by computing unrelated features simultaneously in separate threads. The document provides a simplified example of representing feature engineering equations as a semantic graph and how parallelization could be applied.