Fabrizio Dimino is an AI Research Scientist at Domyn specializing in trustworthy AI for financial services, with experience building agentic systems and predictive ML models. Published research at ICAIF, NeurIPS GenAI in Finance, ICLR FinAI, IEEE ICDM on knowledge graphs, AI governance, red-teaming, and reinforcement learning. His recent research investigates how biases emerge in large language models and how these systems can be made more reliable for high-stakes financial applications.
Building Trustworthy Financial AI: Governance, Bias, and Mechanistic Insights
About this event
Large Language Models are increasingly being integrated into financial workflows, supporting tasks ranging from investment analysis to decision-making assistance. However, these systems can exhibit subtle biases that may impact the quality and fairness of their recommendations. In this talk, I will present our research on positional bias in financial LLMs, a phenomenon where model decisions are influenced by the order in which information is presented rather than the information itself. I will also discuss how mechanistic interpretability techniques can be used to identify the internal circuits responsible for these behaviors and how these findings contribute to AI governance frameworks for financial applications.
Key Highlights
- Understanding positional bias in financial AI systems
- Measuring bias across open-source large language models
- Using mechanistic interpretability to trace model behavior
- Implications for AI governance, risk management, and regulatory compliance
- Best practices for building more trustworthy financial AI systems
Speakers
Launch partner
AgentField is open-source infrastructure for building autonomous software factories and the AI backends that power them. It gives multi-agent systems a single control plane for orchestration, governance, and provenance - so every action your agents take is policy-checked and accountable, with no glue code and no editable logs. Apache 2.0, with SDKs in Python, TypeScript, and Go.
