Eva Kaushik
Eva Kaushik
Data Science and Engineering (DSE)
Eva Kaushik is a PhD student in Data Science & Engineering at the Bredesen Center, conducting her research in the Sensors and Electronics Division at Oak Ridge National Laboratory through the UT-Oak Ridge Innovation Institute. Her work applies statistical modeling and machine learning to power system asset monitoring. She is a recipient of the GATE Fellowship (2026–2027).
She earned her BTech in Information Technology from Guru Gobind Singh Indraprastha University, graduating in the top 3% of her class after a merit-based branch upgrade awarded to the top five students in her freshman cohort, and receiving the Director’s Appreciation for academic performance in her final year.
Before beginning her doctorate, Eva spent three years as a data scientist in industry. At Nestlé, she built and deployed production forecasting and inventory-optimization models on AWS and Azure, delivering nine models against a $30M business case. At DXC Technology, she developed graph neural network models for real-time fraud detection and time-series systems for predictive maintenance, earning four company awards in 2024 for collaboration, mentorship, innovation, and service. She was also technical co-founder of Dexignare, a computer vision startup selected for the Microsoft for Startups Founders Hub.
Her work has appeared in venues published by IEEE, Springer, Wiley, and De Gruyter, and she has presented at ICAAF 2023 in Bucaramanga, Colombia (travel grant), IEEE-ICIET 2023 in India (travel grant), and the Advanced Materials Science World Congress in London as a distinguished speaker. She has served as a reviewer for ICSCPS (Springer, 2024) and on the National Advisory Committee for the IFERP International Conference (2024). Outside her research, Eva serves as Secretary of the UTK Graduate Society of Women Engineers and has organized data science workshops for IEEE-USA and IEEE Region 8. At UTK, she presented her work at EUReCA, traveled to Detroit, Michigan (with a travel grant award), and OSDX 2026 at Oak Ridge National Laboratory.
Research
Substation transformers are now heavily instrumented, and event recordings accumulate faster than anyone can review them, almost none of them labelled. My research addresses the question that must be settled first – how should two such recordings be compared? Because recording is triggered by a threshold crossing, identical events are captured at different offsets, so a naive comparison measures trigger timing as much as event physics. I remove that confound by aligning every pair exactly, then test which definitions of similarity genuinely separate event types. The result is thousands of unlabelled recordings reduced to a small set of interpretable event profiles, moving transformer monitoring from reactive review toward early characterization of asset condition.