Gen AI Model Management in Financial Services: Explainability, Transparency, and Lifecycle Monitoring

Project Overview

The financial services industry is rapidly adopting Generative AI (GenAI), particularly large language models (LLMs), to improve efficiencies and decision-making. This project aims to explore risks, controls, and management principles for GenAI deployment. Topics include explainability techniques, confidence scoring, bias mitigation, stochastic behaviour, input sensitivity, prompt engineering, and future-focused research on agentic AI and domain-specific small language models.

What You Will Do

The student will evaluate interpretability methods like SHAP, ICE, LIME, and counterfactuals to support transparency. They will analyze model output confidence, assess bias and fairness, examine variability and robustness, and investigate prompt engineering strategies. The work will be conducted in collaboration with Nationwide Building Society’s data science teams, providing real-world data and applied impact.

Expected Outcomes

The research will develop methodologies to ensure safe, sound, and compliant use of GenAI in financial services. Deliverables include improved explainability frameworks, bias mitigation tools, dynamic prompt optimization methods, and comparison of LLMs against traditional ML models. The collaboration aims to produce impactful insights contributing to both academia and industry.

Why This Matters

As financial institutions integrate AI deeply into critical processes, ensuring these models are transparent, fair, and well-managed is essential for maintaining trust, meeting regulatory demands, and safeguarding customer interests. This research addresses urgent challenges in AI governance in finance, advancing both knowledge and practice.