Pavan Kotapati

LLM-Powered Automation and GenAI Projects:

• Built an LLM-powered solution using Vertex AI Gemini Pro 2.5 to modernize legacy Python ETL scripts into modular, parameterized versions with YAML-based configurations. Automated 85% of the script transformation process, reducing a 1-year manual project timeline to just 3 months, and saving an estimated $250K+ in developer effort and delivery costs.
• Developed an LLM-based interface using Vertex AI to generate BigQuery SQL from natural language, interpreting table data dictionaries, and executing queries directly in BigQuery. Enabled non-technical teams to self-serve analytics, reducing engineering support hours, and saving approximately $100K annually. Achieved 95% accuracy for single-table queries; expansion to multi-table joins using Dataplex metadata is in progress.

Data Migration and Ingestion Frameworks:

• Coordinated the large-scale strategic migration of data processes from SAS to the GCP-Python ecosystem, minimizing business disruption, and implementing monitoring systems to ensure 99.9% pipeline uptime and scalable operations.
• Designed a modular ingestion framework using YAML-based configurations, with built-in tracking, schema enforcement, DQ checks supported by alerting, and robust logging.
• Collaborated with stakeholders, domain experts to understand business requirements and translate them into efficient and cost-effective data solutions.
• Developed POCs and streamlined GCP data workflows by automating ingestion, transformation, and validation using Python, BigQuery, and DataFlow, enhancing scalability, improving storage efficiency, and reducing processing time by 80%.
• Automated over 7,000 hours of manual data processing annually through ETL pipelines, leading to over $400K in estimated cost savings through reduced manual effort and elimination of legacy SAS licensing fees.

Platform Automation, Machine Learning, and Orchestration:

• Leveraged serverless technologies (Cloud Run, Cloud Functions) and event-driven architectures to enable realtime data processing, integration, and automation at scale.
• Implemented ML workflows using Python, PySpark, and BigQuery ML to generate actionable insights, while optimizing costs through transient DataProc clusters and GCS lifecycle policies.
• Applied best practices in coding, query optimization, and Git-based version control while establishing CI/CD pipelines to automate testing and deployment, ensuring robust, scalable, and efficient production systems.
• Created a Python search macro to scan GCS-stored ETL scripts for target strings. Later, it evolved into a lineage discovery tool, building a job-table dependency matrix to track upstream and downstream jobs.

• Delivered in-depth savings analysis and ad hoc reporting using Python, SQL, and Excel, and built Tableau dashboards that enabled data-driven decisions, contributing to multimillion-dollar savings for clients.
• Collaborated cross-functionally with Product, IT, and Account Executives to translate business requirements into technical solutions, while ensuring data integrity through regular database maintenance and backups.
• Designed and maintained Salesforce dashboards to monitor data quality and call center performance metrics, including response times, call blocking, and abandonment rates.

Thesis: Evaluation of Machine Learning models in Prediction of 5-Year Cancer Survivability.

• Designed ETL pipelines using Python, RStudio, KNIME, and WEKA to implement models with stratified K-fold cross-validation. Applied class balancing techniques to address bias and overfitting, and improve model performance through feature selection, enhancing sensitivity, precision, F1 score, and AUC.

NSCLC Report De-identification:

• Developed an automated pipeline to extract radiology reports from Cerner and load them into REDCap for NLPbased de-identification using the MIST tool. Preprocessed text using Python's NLTK (stemming, lemmatization, stop-word removal, POS tagging) to enhance data quality. Led PII validation to ensure full compliance with HIPAA data privacy standards.

REDCap PHI Integration:

• Developed ETL pipelines for loading PHI data into REDCap via R and its APIs, validated records in the Missouri University Stroke Registry using R scripts, and designed REDCap data collection forms for research and survey workflows.