Research & practice

A walkthrough of the Reins AI monitoring and triage platform showing how domain-specific failures pass standard observability tools undetected, and how severity-based triage routes them to the right people with a full audit trail.

Invited talk at RMAIIG presenting the reliability loop framework (from failure detection through severity classification, synthetic reproduction, and verified repair) for agentic AI systems in regulated industries.

Invited by the EPFL AI Center in Lausanne, Marisa Ferrara Boston presented her framework for building reliable and self-repairing AI systems.

Tutorial at IEEE AI Test 2025 on deploying evaluations for generative AI systems with real-world rigor. Covering system context, measurement and monitoring, agent behavior testing, and evaluation frameworks. Co-taught with Heather Frase, PhD and Sarah Luger, PhD.

Co-author on a structured red teaming exercise at Dana-Farber Cancer Institute testing copyright compliance in a production generative AI system. Participated as a red teamer alongside researchers from MIT, Harvard, Microsoft, MLCommons, and NIH.

Co-author on the first comprehensive industry-standard benchmark for assessing AI-product risk and reliability, developed through an open, cross-sector process at MLCommons.

Designed an organizational framework to guide real-world adaptation of deployed AI agents across high-stakes tasks.

Developed product strategy and funding models to scale AI evaluations and benchmarking in pharmaceutical R&D.

Developed and deployed a framework to evaluate AI agents against human teams in financial services.

Led human feedback and evaluation for a large-scale generative AI imagery tool, ensuring policy alignment and safe public rollout.