The interTwin project is advancing the integration of Digital Twins across scientific domains, focusing on physics and climate research. A key component of this project is itwinai, a Python library designed to streamline scalable AI workflows on High-Performance Computing (HPC) systems. With its unified interface, itwinai simplifies the deployment and optimisation of AI models across leading frameworks for distributed machine learning. The library features tooling for profiling scalability and monitoring GPU utilisation, allowing scientists to better understand and show how well their code is distributed. It also helps to identify inefficiencies, enhancing sustainability and helping to develop greener AI solutions. Recent advancements include support for large-model parallelism and distributed hyperparameter optimization (HPO). By providing a uniform pipeline to run AI workflows easily and intuitively, itwinai lowers the barriers to these complex domains, empowering scientists to achieve reproducible, high-performance results on HPC infrastructure. Through integration with interLink, itwinai facilitates seamless offloading of compute-intensive tasks from cloud to HPC. Validated on diverse use cases in physics and climate research, including collaborations with CMCC, EURAC and CERFACS, itwinai has shown that it has the potential to address challenges in renewable energy, climate modelling, and sustainable development.