EXPERIENCE | pragati

WORK EXPERIENCE

Designed and built an NLP pipeline to classify intracranial hemorrhage (ICH) from over 30k+ radiology reports, leveraging BERT and advancing research into state-of-the-art Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG)
Developed a predictive model on AWS SageMaker to analyze electronic health records (EHR) and identify high-risk, unscreened females for breast cancer, driving proactive health interventions.
Implemented automation procedures to streamline extraction and integration of CMS Hospital Care Compare files and internal data from various vendors, cutting down manual effort of one full-time employee by ∼ 2 weeks per quarterly cycle.

Automation: Led a team of 4 to create ETL pipelines on Azure for data migration verification, reducing ∼ 2 hours of daily manual work.
Customer Segmentation and Engagement: Implemented a clustering model to segment credit card customers based on spending behavior. Through hyperparameter tuning, feature engineering, and evaluation, accomplished 12% increase in customer engagement.
Fraud Detection Model Optimization: Revamped a Credit Card Fraud Classification model, reflected in a significant improvement in AUC-ROC score from 0.82 to 0.89. Contributed to the deployment of the optimized model, enhancing fraud detection capabilities.
Interactive Data Visualizations: Took the initiative to automate & optimize various SQL and Excel-based reports into interactive and real-time Tableau Dashboards. Thereby reducing preprocessing and query time, resulted in savings ∼ 50 hours/month.
Data Modeling: Designed data models and schemas for relational databases, optimizing query performance and storage efficiency.

Improved sales prediction model with diverse ensemble of time series models, leveraging optimal features and new engineered inputs
Captured best inputs from base model features and engineered features like Median of all models and Time of hour
Utilized Support Vector Regressor for aggregation and enhanced predictions by using two models for peak & off-peak hours
Tested the model on past data and achieved an average of 1% improvement over various accuracy metrics

• Tutored high school/UG level students as an independent contractor on the Chegg platform, achieving a 95% satisfaction rate.

• Experience in teaching over 60+ students and conducting 80+ lessons through the platform.

• Taught students SQL, Database Management, Python/C++ Programming, and guided them in solving projects and assignments