The Duke–BNY capstone team’s paper, “FedSight AI: Multi-Agent System Architecture for Federal Funds Target Rate Prediction,” has been accepted to the NeurIPS 2025 Workshop on Generative AI in Finance, a significant recognition of the team’s research and innovation.
NeurIPS (Conference on Neural Information Processing Systems) is one of the world’s premier gatherings for researchers in artificial intelligence and machine learning. Known for its rigorous peer review and highly competitive acceptance rates, NeurIPS showcases groundbreaking research that shapes the future of AI. Its workshops bring together leading academics and industry experts to explore frontier topics and practical applications of AI. The Generative AI in Finance Workshop focuses on how cutting-edge generative methods can advance modeling, decision-making, and innovation in financial systems.
The paper highlights the outcomes of a collaboration between Duke University’s Master in Interdisciplinary Data Science (MIDS) program and BNY (Bank of New York Mellon), showcasing the program’s interdisciplinary and applied approach. Duke MIDS trains students to integrate data science, social science, and domain expertise to solve complex, real-world problems.
Tianji Rao ’25
“As a former Duke MIDS student and member of the 2024–2025 Duke–BNY Capstone team, I am deeply grateful for the program’s commitment to interdisciplinary collaboration, ethical AI development, and professional readiness. The capstone experience equipped me with the technical and professional skill set essential for success in applied AI, directly paving the way for my transition to a full-time role at BNY’s AI Hub,” said Tianji Rao ’25, Senior Associate, AI/ML Software Engineer at BNY.
The project (completed as part of the Duke MIDS Capstone, a yearlong, team-based partnership with an external organization) allowed students to apply classroom learning to a real-world AI challenge. Capstone teams collaborate with clients from industry, government, and academia to deliver data-driven solutions, blending technical rigor with professional experience.
FedSight AI employs generative AI and a multi-agent system to simulate Federal Open Market Committee (FOMC) deliberations and forecast monetary policy decisions. The study found that structured agent interactions, each representing distinct decision-making personalities, can emulate the collective reasoning process of human committees. This work illustrates how responsibly designed AI systems can enhance understanding of complex economic dynamics while complementing human judgment. This collaboration provided a unique opportunity to apply classroom learning to high-stakes, institutionally relevant challenges, bridging academic insight with practical implementation under the mentorship of Professor David Ye.
Adler Viton ’25
Adler Viton ’25, Senior Associate, AI Hub at BNY explained, “One of the main reasons I chose Duke MIDS was its rigorous capstone program—and my collaboration with BNY far exceeded expectations. Spending a year tackling a real-world AI and finance challenge alongside industry experts was an incredible experience. We built lasting relationships, made meaningful discoveries, and gained skills that prepared me to join BNY’s AI Hub after graduation.”
Both students expressed their sincere thanks to Professor David Ye, Professor Nick Eubank, and the Duke MIDS faculty for their mentorship and for fostering a program that bridges innovation, scholarship, and real-world impact.
—————
Collaboration in Action! Our recent MIDS Capstone Project was a true team effort, bringing together Duke MIDS faculty, students, alumni, and our partners at BDNY. This collaboration showcased the power of connecting classroom learning with real-world data challenges. A huge thank you to everyone who contributed their time, expertise, and insights to make this project a success!
Capstone Team Members: David Ye, Tianji Rao, Jeremy Tan, XiYue (Vivian) Zhang, Adler Viton, Yuhan Hou
Bank of New York AI Lab Team: Abhishek Kodi, Sanjana Dulam, Aditya Paul