Gaurab Pokharel held a bachelor’s degree in computer science (with high honors) and mathematics from Oberlin College and an M.S. in computer science, with a machine learning concentration from George Mason University when he began searching Ph.D. programs in computer science.

His research focus on how AI, human judgment, and randomness interact in high-stakes systems that allocate scarce resources to people -- particularly homelessness services and housing -- led him to his advisor, Sanmay Das, at the Sanghani Center.

“Dr. Das’s  research at the intersection of AI, economics, and social impact aligns very closely with my interests,” he said. “And the Sanghani Center stood out because it brings together researchers working on both the technical foundations and the real-world applications of artificial intelligence. That combination is exactly the environment I wanted for my doctoral work.” 

In describing his research, Pokharel gave this example: When a city has limited shelter beds or housing vouchers, caseworkers have to decide who gets help first. He studies whether artificial intelligence tools like large language models (LLMs) can reliably make those prioritization decisions.

“I've found that current LLMs are surprisingly inconsistent and unreliable in this setting," he said.

Pokharel also uses machine learning to understand what makes experienced caseworkers good at their jobs and it turns out, he said, that their "discretionary" decisions, the ones that deviate from heuristical rules, tend to be strategic and welfare-improving, not arbitrary.

“Together, this work makes the case that we need to design AI systems that assist human professionals rather than replace them,” he said. 

He has also done work on mathematical models of fairness in repeated selection processes, studying how feedback loops in systems like admissions or hiring can entrench inequality over time even when the rules appear fair at any single point.

His research focus stems from his undergraduate years when he built a convolutional neural network to recognize emotions from facial images. 

“The model performed well on the benchmark data, but when I tested it on my own face, it couldn't recognize me. As a then second-year student, I dug into why and discovered that the training dataset was almost entirely Caucasian faces. The bias in the data had crept directly into the model. That experience made me realize that even well-performing and well-meaning AI systems can quietly fail for the people they're supposed to serve, and that technical performance metrics alone don't tell the whole story,” he said.

From there, he started reading the fairness, accountability, and transparency literature, and his interests deepened from algorithmic bias into the broader question of how AI interacts with human judgment in high-stakes settings: who gets resources, how those decisions are made, and whether the systems we build actually work for everyone. 

“That trajectory eventually led me to my current research on AI in social service allocation, where the consequences of getting it wrong are very real and very consequential,” he said. 

Pokharel’s collaborative published work includes:

·       "Fixed Points and Stochastic Meritocracies: A Long-Term Perspective," ACM FAccT Conference 2026 coming up June 25-28.

·       "Beyond Automation: Understanding Fairness, Ethics, and Human Discretion in AI-driven Societal Decisions," in proceedings of the AAAI/ACM Conference on AI, Ethics, and Society 2025 

·       "Street-Level AI: Are Large Language Models Ready for Real-World Judgments?" in proceedings of the AAAI/ACM Conference on AI, Ethics, and Society 2025 

·       "Discretionary Trees: Understanding Street-Level Bureaucracy via Machine Learning," in proceedings of AAAI Conference on Artificial Intelligence 2024  (AI for Social Impact track) 

“My research spans machine learning, economics, and public policy, and the Sanghani Center makes it easy to find collaborators and have conversations that cut across those boundaries,” he said.

Projected to graduate in Spring 2027, Pokheral said he’d like to pursue a postdoctoral research position or a tenure-track faculty position at a research university where he can continue working on AI for social impact developing evaluation frameworks for AI in high-stakes domains, modeling human-AI collaboration in public services, and advancing the mathematics of dynamic resource allocation.