Small and cheap electronics, less expensive launching capabilities, and the advent of application-specific integrated circuits make constellations of capable small satellites conducting missions in an aggregated fashion. This enables constellations to be resilient in a congested and contested environment. Distributed satellite systems can introduce significant improvements to most missions because space missions are constrained by orbital motion and uplink/downlink capacity. In addition, intelligent computational capabilities reduce the necessary computational load on the spacecraft and optimise the use of resources. Artificial Intelligence (AI) on edge is needed to optimise data processing at the end node. Small satellites can be empowered by AI-capable devices to map raw sensor readings to actionable decisions in near-real-time. However, AI model development for space applications isn't well studied in the literature. Therefore, the authors introduce the feasibility of leveraging data-centric AI approaches to build robust and reliable models that behave as expected beyond training data distributions for the first time in the literature. The developed platform can leverage small and noisy data and keep the model architectures fixed to facilitate space machine learning operations. This work is significant for any space mission requiring cutting-edge machine intelligence.