Recent technological advancements and infrastructure upgrades triggered significant changes in modern synchrotron beamlines. Consequently, experiments are evolving to be more data-intensive and data-driven, increasingly relying on online data analysis for resource efficiency. Machine-learning (ML) based approaches play a crucial role in enabling real-time decision-making through online data analysis and closed-loop feedback applications. In line with advancements in ML-based analysis of X-ray reflectometry, we present both the underlying ML models and their integration into closed-loop operations during experiments. Concerning infrastructure, we rely on widespread ML frameworks, deployed on VISA and coupled via TANGO to the ESRF-developed control system BLISS.Concerning the ML-model, our approach involves incorporating prior knowledge to regularize the training process across broader parameter spaces. This method demonstrates effectiveness across various scenarios, utilizing physics-inspired parametrization of scattering length density profiles extracted from x-ray reflectivity measurements. By integrating prior knowledge, we improve training dynamics and tackle the underdetermined nature of the underlying inverse problem. We illustrate the scalability of our approach by illustrating its use by applying it to an N-layer periodic multilayer model with more than 15 open parameters.