Hire Srivani I. - AI Researcher

* Architected and optimized big data pipelines, processing gigabytes of ionospheric data daily. * Implemented dynamic pipelines with storage event triggers to automate ingestion and real-time processing of data, reducing manual intervention by 85%. * Developed a PyTorch spatio-temporal Transformer model to predict patterns, achieving 99% accuracy with relevant influential parameters and 87% without. * Applied matrix completion method to fill gaps in GPS data while considering the spatial (latitude & longitude) and temporal factors on the global scale. * Optimized the high-resolution data (181x361x105,408 dimensions) for model training using parallel computing on 36 A100 GPUS, reducing the processing time by 62%. * Designed feature engineering techniques to enhance model performance by leveraging spatial-temporal patterns. * Utilized explainability tools such as SHAP to measure feature importance in the model’s decisions. * LWPC simulation to extract amplitude and phase from the raw data in Linux and Julia versions * Processed HTTP data from 143 websites (3 features per packet: time, direction, size) with 2000 packets to detect new labels using an ensemble Machine Learning algorithm.

* Built and optimized a predictive model with NASA’s Sherlock Data Warehouse, achieving 97% accuracy across 100s of flight data points * Developed a robust data pipeline integrating data from various resources with a streamlined dataflow for efficient processing. * Researched, designed, and implemented solutions to enhance pre-trained models, addressing critical challenges in development, evaluation, and deployment. * Evaluated metrics such as demographic parity, equalized odds, or disparate impact to assess regional fairness. * Identified the root cause of bias (i.e. data bias) and applied solutions using weighing techniques and cost-sensitive learning.