Benchmarking trust: Engineering AI safety pipelines for financial services and life sciences

Abstract

As Large Language Models (LLMs) move from experimental sandboxes to production environments in regulated industries, the margin for error is non-existent. In sectors like financial services and life sciences, “hallucinations” are not just quirks—they are compliance risks and safety hazards. This session explores the engineering challenges of operationalizing AI safety.

Drawing on current ML research and my background in quantitative finance and life science research, I will discuss methodologies for designing domain-relevant AI safety benchmarks.

Attendees will learn practical strategies for integrating these safety checks into MLOps pipelines, ensuring that LLM deployments remain reliable, compliant, and valuable in compliance sensitive environments.

Bio

Shantanu holds a Ph.D. in Biochemistry from UNC Chapel Hill and a B.Tech. in Computer Science from IIT Kanpur. His diverse professional experience includes co-founding tech startups (mortgagetech STEM Lending) and driving innovative R&D projects in life sciences, financial services and AI for global institutions like Goldman Sachs, Citigroup, and GE Research.

Shantanu is interested in working in AI applications for financial services and life sciences. He is involved in building global communities for life sciences research, advocating for the safe and impactful adoption of machine learning in regulated sectors.