Alina Mirét Shah

amshah1022.github.io

Cornell University · B.A. Computer Science

Building reproducible evaluation and interpretability systems for transparent, reliable AI.


Research Statement

Triangulating AI Reliability: Evidence, Mechanisms, and Human Judgment
Outlines my direction toward a reliability layer for AI safety, integrating evidence-grounded evaluation, mechanistic interpretability, and human/rubric adjudication.


Current Projects


Skills & Focus


Affiliations

Conducting multi-lab research on evaluation, interpretability, and human-aligned reliability systems.

Cornell University

Long-Term AI Safety Research Lab (Lionel Levine) — Interpretability reliability and meta-probing infrastructure; developing evaluation methods for model understanding tools.

Future of Learning Lab (René Kizilcec) — Rubric-anchored evaluation pipelines and dialogic feedback systems for reliable AI in education.

AI & Robotics Lab (Angelique Taylor) — Systematic review of multi-agent reinforcement learning applications in healthcare.

Culture & Computation Lab (David Mimno & Mathew Wilkens) — Representation alignment and literary corpus analysis using multimodal embeddings.

Former: Horizon Therapeutics · Stripe · Discovery Partners Institute (Alvin Chin)


Documents



About Me

I’m a researcher focused on developing auditable, evidence-based evaluation systems that make model reliability measurable. My goal is to bridge mechanistic interpretability, evidence-grounded NLP, and human-aligned evaluation to build transparent AI pipelines ready for oversight and safety-critical domains.


Last updated: October 2025

Site maintained by Alina M. Shah · amshah1022 @ gmail.com