Krishna Srujan Vaddiparthi

About Krishna Srujan Vaddiparthi

I’m a Data Scientist working at the intersection of engineering systems and machine…

Experience

1. Data Scientist Co-op

   ITT Goulds Pumps

   Jan 2025 — Present · Seneca Falls, NY, US

   Hydraulic Engineering Dept. – Pump Performance Modeling Project- Built and scaled ensemble pipelines for predicting 3 key pump metrics, extending legacy baselines with stratified bootstrapping and LOOCV- Developed AdaBoost and GBDT ensembles using modular pipelines (scaling, balanced splits, LOOCV, stratified bootstraps), benchmarked on a fixed holdout set- Boosted critical target’s R² from 30% to 82% via refined training strategy, feature engineering, and ensemble tuning, validated with CI ranges, FN/TP%, and generalization gap metrics- Tuned a 30-model GBDT ensemble for the critical target using MSE, MAPE, R², and holdout metrics to select interpretable, high-performing subsets- Modularized codebase with reusable functions for bootstrapping, LOOCV, ensemble predictions, & CI evaluation- Exploring XGBoost and PINNs with new features to boost generalizability across diverse pump setups.iAlert’s Generic Diagnostic ML model project- Built a multi-class classification pipeline for fault detection on rotating equipment, capturing over Parquet files (each with 4,096 FFT data points) from various pumps (5HP–125HP). Overcame sensor connectivity with asynchronous BLE calls, error-retry logic, and data-validation checks, ensuring reliable Python-based ETL- Used magnetic flux data to auto-detect run speed, dynamically choosing between 1600-bin (low-res) or 2496-bin (high-res) FFT feature extraction. Implemented two logistic regression models (low-res vs. high-res) to compare predictive performance across multiple fault classes- Raised test accuracy from 18% to 64% when tested on a new 125HP pump at 50Hz—outperforming a physics-based model (45%) with limited resources. Maintained 74% training accuracy and minimal overfitting, showing generalizability- Performed in-depth EDA (label encoding, scaling, confusion matrices) to refine classification pipelines, and communicated findings across teams, driving data-driven decisions for predictive maintenance.

Education

• BITS Pilani, Hyderabad Campus

  Bachelor's degree, Mechanical Engineering

• Rochester Institute of Technology

  Master of Science - MS, Data Science