R, Python, Sql, Agile issue tracking.

• Wrote code and queries, to analyze customer usage data, and develop ML & DL models for anomaly detection and time series classification.
• Tsfresh time series (TS) features were extracted and dimension reduction techniques; and different ML classifiers (Random Forest, SVM, QDA & Neural Net) were used to classify anomalous time series.
• Feature engineering : Significant features were obtained using CCA, PCA, TSNE & Baruta methods.
• Change point analysis was used to identify anomalous premises, and ggplot visualizations done for 2 million accounts.
• Computations were implemented across threads in parallel to speed up computations using shared memory

Rapid Alloys project: Code in R, Python & IBM SPSS.

• Different 9Cr Steel test data of inhouse and those from Japan were analyzed : to develop data analytic tools for rapid development of new alloys at NETL in collaboration with Case Western University.
• Pairwise correlations were done as part of exploratory analysis. Variables of non-zero variance & frequent occurrences were used to predict strength, ductility and other mechanical properties.
• Multi-variate and stepwise regression were used to determine rank order of variables in explaining a mechanical property.
• Random forest recursive partitioning technique was used to explore expanded composition space to predict property of new alloy from existing test data.
• Different techniques were used to delineate different classes of alloys i.e. ANOVA, and Tukey HSD tests on mean of samples, principle components, stochastic embedding (t-SNE), hierarchical and medoid, kNN clustering have been used to visualize different classes of alloys in 2 and 3 dimensions

Paper: https://doi.org/10.1016/j.commatsci.2019.03.015

NIH Multimodal Imaging Fellow Award (2012-2013), RF simulation using threaded C code on Linux and Matlab.

• Designed and prototyped components used CAD (Solid works) and workshop for prototyping.
• Model extraction from CAD models, and development of multiple anatomically detailed models and evaluation of MR coils simulations using full wave Maxwell's equations
• Independent Transmit (Tx) & Receive (Rx) coil arrays are used to tackle (electric and magnetic) field in homogeneity that arise at ultra-high field MRI. An important part of my thesis work was bench marking of coil performance for efficient and safe use in-vivo.
• The arrays were tested for reproducibility, reliability and safe usage across multiple studies. EM/RF Finite Difference Time Domain simulations of the Tx and composite of five head models were used to optimize parameters, to obtain homogenize whole brain excitation with low RF absorption.

Papers: https://doi.org/10.1371/journal.pone.0209663 and https://doi.org/10.1002/jmri.24152