I develop AI systems for molecular discovery, multi-omics analysis, and precision medicine.
I'm an AI ML scientist working at the intersection of machine learning and computational biology, designing generative models for proteins and RNAs, predicting protein behavior with language models, and building diagnostic AI for cardiac health. My goal is to make drug discovery faster, cheaper, and more accessible.
University of Kansas.
Selected to serve on the Program Committee for leading AI and Computational Biology conferences, contributing to the peer-review process of cutting-edge research. IEEE International Conference on Bioinformatics and Biomedicine (BIBM).
KU graduate research assistant publishes breakthrough studies across AI, healthcare, and sustainability University of Kansas.
Our paper on privacy-preserving ECG diagnostics is published in Springer Nature Scientific Reports.
NbBayesLM paper accepted at Frontiers in Bioinformatics — Bayesian uncertainty quantification for nanobody thermostability.
CIgFlow preprint posted — antigen-conditioned antibody generation via conditional flow matching.
Protein-to-metal binding prediction paper accepted in IEEE Transactions on Computational Biology and Bioinformatics.
Bloom Filter IC authentication framework posted on arXiv.
Presented MetaLLM at IWBBIO 2023 in Granada, Spain.
Two journal papers in 2026 — Scientific Reports and Frontiers in Bioinformatics.
Research supported by the American Heart Association (AHA) — collaborative work on cardiac AI and antibody design.
Appointed Visiting Researcher at KU Medical Center, School of Medicine for ECG AI collaboration with cardiologists.
Best Paper Award — IEEE IEMCON 2020 for "Graph Theory for Dimensionality Reduction: Prognosticating Parkinson's."
Joined the University of Kansas as a Graduate Research Assistant in the EECS department.
Shipped production ML systems at BJIT Group and Inside Maps — face recognition, real-time detection, and forecasting at scale.
I'm a machine learning researcher with a focus on protein design, drug discovery, and computational biology. My day-to-day involves training and fine-tuning large protein language models, building generative frameworks for antibody engineering, and developing clinical AI tools for cardiac diagnostics — all in close collaboration with biologists and physicians.
Before starting my PhD, I spent two years as an AI ML software engineer building production ML systems — face recognition, real-time object detection, and forecasting pipelines deployed on AWS at scale.
I'm part of an AHA-funded project and hold a visiting researcher position at the KU Medical Center, where I work on privacy-preserving ECG analysis with the School of Medicine.
"I believe the next breakthrough therapies will emerge where artificial intelligence meets biology—turning complex molecular data into actionable insights for human health."
Presenting research in conference
CIgFlow is a generative framework that learns a continuous mapping from joint antibody-antigen embeddings to novel antibody sequences. It combines Conditional Flow Matching with a GPT-2-style autoregressive decoder, protein language model embeddings, and classifier-free guidance — generating antigen-specific antibodies while maintaining humanness and developability. The ODE-based latent sampler allows diverse exploration of sequence space without sacrificing quality.
Predicts nanobody melting temperature (Tm) using ESM-2 and AbLang embeddings as input to a Bayesian deep learning framework that also encodes physicochemical features as priors. Evaluated on 10,000+ nanobody sequences: MAE of 1.89°C and R² of 0.67. The model decomposes epistemic and aleatoric uncertainty, letting researchers prioritize which candidates are actually worth synthesizing.
GPCRs are the target of ~35% of all approved drugs, but classifying them at scale is computationally expensive. GPCR-SLM uses knowledge distillation to compress large protein LM knowledge into a lightweight transformer that classifies 86 GPCR families at 99% accuracy while being 33.5× faster. Validated on clinical variants, metagenomic sequences, and tumor-derived data. Also supports open-set recognition for novel families.
A machine learning pipeline that combines SMILES representations, molecular fingerprints, and physicochemical descriptors with ChemBERT to predict drug-induced toxicity. Achieves 92% accuracy on cardiotoxicity prediction, significantly outperforming traditional QSAR baselines. Built as part of parameter-efficient deep learning work for computational drug discovery.
A CNN-VAE architecture for analyzing 12-lead ECG signals that directly addresses demographic and acquisition bias — a known problem in cardiac AI that causes models to perform worse on underrepresented populations. Achieves 88% sensitivity and 94% specificity. Privacy-preserving by design. Published in Springer Nature Scientific Reports, 2026.
I'm always interested in research collaborations, visiting opportunities, and conversations about computational biology, drug discovery, or biomedical AI. If you're working on something interesting — even tangentially related — feel free to reach out.
I typically reply within 24–48 hours.