The standardization of intrapartum care, as promoted by the World Health Organization’s Labour Care Guide (LCG), relies on the consistent monitoring of maternal and fetal well-being. A critical component of this process is the assessment of fetal head progression using intrapartum ultrasound, specifically by measuring the Angle of Progression (AoP). However, manual landmark annotation for AoP calculation is time-consuming and prone to inter-observer variability, creating a bottleneck in clinical workflows. To address this, we propose an automated approach for fetal biometry. The core challenge in developing such a deep learning-based solution is the complex and computationally expensive task of optimizing neural network architecture and hyperparameters. In this work, we introduce a novel framework based on a Generative Reward Machine (GRM) to create an autonomous agent that intelligently and efficiently navigates the joint space of batch size and hyperparameters. This method accelerates the development of robust models for automated landmark detection, aiming to enhance diagnostic consistency and support the implementation of safer, evidence-based intrapartum care.