My research will explore how artificial intelligence methods such as machine learning, natural language processing (NLP), and speech processing, can be developed and applied to support healthcare decisionmaking in a reliable, interpretable, and clinically meaningful way. Rather than focusing on a single predefined application, the research will investigate shared challenges that arise across healthcare AI tasks, especially when working with complex, multimodal, and imperfect data. Key questions guiding this work include: How can speech and text data (e.g. clinical notes, patient-clinician interactions, or patient-reported information) be effectively integrated with structured clinical data? How can AI models remain robust, transparent, and fair when applied to heterogeneous patient populations and real-world clinical settings? How should such models be evaluated to reflect clinical utility rather than purely technical performance? Across the first two years, the research will combine methodological exploration, applied experimentation, and close engagement with healthcare data, allowing the direction of the work to evolve as new insights emerge from both modelling and clinical contexts. The research themes and exploratory work packages are highlighted below: Data and Problem Exploration Initial work will focus on understanding the characteristics and limitations of healthcare data, including structured records, clinical text, and speech-based data. Exploratory analyses and baseline models will be used to identify clinically relevant problem settings (e.g. risk stratification, early detection, or monitoring) and to surface challenges such as missing data, bias, and dataset shift. Speech and Language-Based AI for Healthcare A core research theme will investigate NLP and speech processing methods for healthcare applications, including representation learning from clinical text and spoken language. This will involve exploring approaches that capture clinically meaningful information while addressing issues such as noise, domain shift, and limited labelled data. AI Methods for Reliability and Interpretability The research will explore AI techniques that prioritise model reliability, including interpretability, uncertainty estimation, and robustness. Particular attention will be paid to understanding how speech- and language-based models behave in clinical contexts and how their outputs can be communicated clearly to end users.