A design process often follows several sequential phases. The main challenge is to obtain a solution that meets the project objectives, requirements and expectations while taking into account factors from all phases of the design. This usually requires extensive manual adjustments using various feasibility and performance analyses, which can be time-consuming and costly. Machine learned generative design is an alternative approach in which a system is developed that replaces the results of the analyses and combines them into a single surrogate model. This enables the predictions of overall solutions that meet all project requirements and objectives. The effectiveness of this approach was tested on the real-life example of the design of the New Robotic Telescope observatory. This involved a clamshell-type enclosure characterised by a movable roof and the associated design states. With our approach, we captured the results of each state using machine learning models and combined them into a single surrogate model that allowed us to identify solutions that satisfied the conditions and objectives of all states. The results not only demonstrate the success of our method, but also highlight its potential to improve the efficiency of the design process.