Unlimited access to clean, sustainable and renewable source of energy seems to be a dream scenario, however with an assistance of ML (machine learning) support the experimental work which as a matter of fact reveals the right descriptors may gain a powerful tool allowing to omit a time consuming trials path. However, precise definition of descriptors will require an identification of the bottleneck materials parameters, features and processes, thus different architectures and working arrangements will be shown and discussed in this presentation in order to test their usfulness for prediction of novel and better arrangements for hydrogen and other carbon based solar fuels production. When a suitable material is identified, then there are numerous strategies to overcome potential bottlenecks, including doping, nanostructuring, surface modification, vacances optimisation or addition of co-catalysts, thus a game changer is a precise screening of possible predictors in order to identify the most performance determinant factor. Therfore, over 400 tailor-made for PEC reduction process samples based on the semiconductor oxides have been prepared and measured in order to bulid a pre-base for ML tests. In this presentation we will discuss the strengths and weaknesses of different architectures and working arrangements for hydrogen and carbon-based solar fuels production; different materials and working systems, their compositions, geometry, properties will be discussed in view of their impact on the final prediction.