I am a Master's student in Electrical and Computer Engineering at the University of Michigan, Ann Arbor. My research interests lie in developing and applying novel machine learning methods, physics-based computational models, and optimization techniques to real-world systems.
I work with Prof. Ziyou Song at the Electric Vehicle Center, where I have built physics-based electrochemical impedance models for Li-ion battery characterization using differential evolution optimization, and developed attention-enhanced sequence-to-sequence models for battery aging prediction. I also work with Prof. Raj Rao Nadakuditi, where I am translating and understanding a research codebase for electromagnetic wave scattering simulations from MATLAB into Julia and Python.
Before joining UMich, I completed my B.Tech in Electrical Engineering from IIT (ISM) Dhanbad, where I developed a strong foundation in machine learning and computer vision. I am actively looking for summer research opportunities.
Developing ML models that incorporate physical priors and domain knowledge for scientific applications, including sequence-to-sequence models for time-series prediction
Building physics-informed models that couple first-principles equations with data-driven approaches for interpretable parameter extraction and system characterization
Applying evolutionary and gradient-based optimization to non-convex, high-dimensional problems arising in scientific and engineering systems
Applying deep learning models for image segmentation, 3D reconstruction, and image processing in scientific and engineering applications
Working on hybrid physics-based EIS model paper draft for SEI characterization in Li-ion batteries (with Prof. Ziyou Song)
Started working with Prof. Raj Rao Nadakuditi on translating and understanding electromagnetic scattering simulation codebase
Joined the Electric Vehicle Center (Prof. Ziyou Song) working on physics-based battery modeling and ML-based aging prediction
Started MS in ECE at University of Michigan
University of Michigan — Prof. Raj Rao Nadakuditi
Translating the CyScat electromagnetic scattering simulation codebase from MATLAB to Julia and Python. Working with scattering matrix (S-matrix) methods for periodic arrays of cylinders, generating wavefront simulations, and understanding eigen-channel analysis via SVD of scattering matrices (S11 for reflection, S21 for transmission) to identify optimal wavefronts that maximize transmission or reflection.
Electric Vehicle Center, University of Michigan — Prof. Ziyou Song
Built a physics-based EIS modeling framework combining equivalent circuit models (SEI/CEI layers) with DRT analysis; extracted impedance parameters tracking SEI evolution from BOL to EOL across cells, SOH, and SOC levels via Differential Evolution optimization. Developed attention-enhanced Seq2Seq models for battery aging prediction, achieving 2–5% MAPE on full capacity-fade trajectories from the first 30% of cycling data. Designed accelerated testing protocol framework converting real-world driving profiles to battery power demands.
Carnegie Mellon University
Worked with Dr. Arun Balajee Vasudevan on content-agnostic deepfake speech detection. Developed benchmark datasets using multiple TTS models and trained audio-language detection models. Paper submitted to Interspeech 2026 (under review).
Mowito
Implemented perceptual-hashing based methods for large-scale dataset deduplication and built production-ready dataset pipelines. Developed Multi-head Mask R-CNN models for instance segmentation tasks in warehouse environments, and built order-completion time estimation and clustering algorithms for warehouse optimization.
Hybrid physics-based SEI impedance model with degradation-constrained differential evolution optimization, achieving 1.7% mean error across 6 cells and 117 EIS spectra
Learn More →Translation of EM scattering simulation codebase from MATLAB to Julia/Python, with wavefront optimization via scattering matrix eigen-channel analysis
Learn More →Attention-enhanced sequence-to-sequence models for battery capacity-fade trajectory prediction from early cycling data, with 2–5% MAPE
Learn More →Matrix Methods for Machine Learning and Signal Processing
Probability and Random Processes
