Pernille Matthews

Research and development of methods for understanding and validating machine-learning behaviour.

My work focuses on analysing how models make decisions, identifying unreliable predictions, and creating techniques that allow organisations to trust AI systems in practice. This includes designing evaluation frameworks, failure-mode analysis, and translating research algorithms into implementable software pipelines.

The research is applied across classification, clustering and anomaly detection and targets real-world deployment challenges in addition to theoretical settings.

Founded a technical consultancy focused on machine-learning systems and data-driven applications.

I work with organisations to design, validate and stabilise AI solutions before and during production deployment. My work typically involves analysing model behaviour, improving reliability, and translating research methods into robust software systems.

Example engagements include:

• Healthcare prediction models (COPD deterioration risk) in collaboration with Region Sjælland

• Development of research and educational platforms using Django and Python

• Design and implementation of data pipelines and machine-learning solutions for industry

• Architecture and redesign of internal planning and scheduling software

The focus of my consultancy is reducing technical risk in AI projects and ensuring solutions can be maintained and trusted in real-world use.

Independent research position following my MSc, focusing on explainable machine learning methods.

Worked on analysing model decision behaviour and implementing prototype algorithms for understanding predictions in classification models. The role served as a transition into doctoral research and involved designing experiments, evaluating methods, and communicating technical findings to both technical and non-technical audiences.

Additionally contributed to academic events and supported teaching within data science to make complex ML concepts accessible in practice.