Analysing and understanding movement is critical across disciplines, including neuroscience, as it offers insights into neural mechanisms underlying motor control and behaviour. However, the current approach to movement analysis often results in numerous small, domain-specific models that struggle to capture shared movement features, limiting their real-world applicability. In this talk, I will showcase data-driven approaches that leverage pre-trained models, such as vision transformers, to optimize small movement datasets and highlight promising results from applying these methods to automated video-based risk assessment for Cerebral Palsy in infants. Lastly, I will discuss how foundation models present new opportunities for neuroscience by enabling richer behavioural analysis and paving the way for applications like high-bandwidth neuromotor interfaces.