Research Collaborator — Molecular Generative Modeling
Old Dominion University
Collaborating on Phase I of PolymerGPT, a decoder-only GPT-style transformer trained from scratch on polymer SMILES representations, using multi-GPU distributed training on Google Cloud (Vertex AI); manuscript under review, co-first author.
Achieved Validity/Uniqueness/Novelty (VUN) scores above 95% across all data subsets, surpassing existing literature benchmarks.
Optimized the TransPolymer property-prediction model using Ray Tune (Bayesian optimization + Hyperband scheduling).
Lead Researcher
Netramind Innovations
Leading and mentoring a team of researchers and engineers; drove adoption of a QMS aligned with GCP and FDA 21 CFR Part 11 standards.
Owned full-stack development and deployment of 3 production, revenue-generating clinical AI applications on the Heidelberg Engineering AppWay ecosystem within one year — CVI, NMI-ORA, and NMI-AMD-SC — spanning model training through Azure deployment and HE integration.
NMI-ORA is central to 3 completed clinical trials plus an ongoing multi-center study; NMI-AMD-SC is deployed across a 10+ optometrist network in Pennsylvania.
Led a multi-team study proving that Heidelberg’s device-reported image quality (Q-score) is not a reliable predictor of segmentation performance, across 5,000+ OCT B-scans, and introduced a novel task-specific quality metric (preprint on medRxiv, under review at Translational Vision Science & Technology).
Researcher (Fellow, Senior and Professional)
University of Pittsburgh / UPMC
Solo-developed and independently implemented a label-free deep learning framework for retinal layer segmentation on Cirrus OCT — now deployed on UPMC MyApps and used clinically across the health network.
This label-free approach directly enabled Netramind Innovations to build and commercialize NMI-ORA.
Collaboratively deployed the framework to production with the UPMC enterprise team, serving as the point of contact (POC) for the integration — including an ONNX pipeline refactor that achieved an 8x speedup (12 min → 90 sec per volume).
Presented 11 posters and served as Session Moderator for the Retina in AI track at ARVO 2025; contributed to 14 additional posters.
Graduate Assistant (Teaching and Research)
University of North Carolina at Greensboro
Researched domain adaptation techniques and structurally informed deep learning models for robustness across varying image characteristics.
Managed 3+ courses per semester (Deep Learning, Image Processing, Algorithm Analysis); Summer 2023 instructor for Elementary Data Structures & Algorithm Analysis, 93% student approval rating.
AI/ML R&D Software Engineer
Carelon Global Solutions (formerly Legato Health Technologies, Elevance Health)
Built a ticket-bucketing text-classification system using embeddings that automated manual triage, saving $100K annually.
Designed AI-driven telehealth features for clinician–patient communication with disabilities; built CNN-based multimodal models for patient readmission risk.
Research Fellow
LV Prasad Eye Institute
Built an intra-ocular lens (IOL) power prediction model (Random Forest) matching industry-standard formulas.
Built a custom OCR pipeline to digitize and analyze 60,000 cataract-surgery records — the first large-scale presentation of real-world cataract data at this scale in India.
Data Scientist
Cyient Insights
Built proof-of-concept failure-prediction and anomaly-detection models under class-imbalanced settings for international clients.
Built Tableau dashboards for business intelligence and operational insights.
Education
M.S. in Computer Science
University of North Carolina at Greensboro
Thesis: Exploring Explainability of Hypergraph Neural Networks using Mixed-Integer Programming.
M.Tech. in Electrical Engineering (Communication & Signal Processing)
Indian Institute of Technology Hyderabad
Thesis: Towards True 3D Display — Technological Feasibility of a Flatscreen Solution.