Igor Melnyk

I am an Applied Researcher at Capital One, where I focus on developing AI Foundation models for tabular and multimodal financial data. My recent work centers on transformer-based architectures for modeling customer behavior using structured financial datasets, with an emphasis on decision agents, event sequence modeling, and graph-based representations of customer interactions.
Before joining Capital One, I was a Research Staff Member at IBM Research, where I worked in Machine Learning, Natural Language Processing, Computer Vision, and Biomedical AI. My early work at IBM included developing deep generative models for text, images, and knowledge graphs, with a focus on modality transfer tasks such as image captioning, semantic text rewriting, and text-to-graph generation. With a background in Probabilistic Graphical Models, I also explored novel methods for anomaly detection in multivariate time-series data, contributing to applied projects in financial and industrial domains.
I have also contributed to research in Computational Biology, particularly in Protein and Antibody Design. My work leveraged pre-trained foundational models like AlphaFold to develop inverse folding techniques and conditional structure generation using denoising diffusion models. I also investigated model reprogramming approaches to adapt Large Language Models for antibody sequence optimization to help accelerate drug discovery.
Earlier in my academic career, I specialized in Robotics and Bayesian time-series modeling. I worked on collaborative localization for multi-robot systems and later focused on modeling high-dimensional, heterogeneous data streams using Dynamic Bayesian Networks. My doctoral research involved developing efficient algorithms for learning and inference in autoregressive and Hidden Markov Model frameworks for anomaly detection.
I hold a Ph.D. in Computer Science and Engineering from the University of Minnesota, and an M.S. in Computer Science from the University of Colorado Boulder.