Deepeka Gurunathan

Data Analyst Intern

Texas Education Agency · Texas, USA

• Analyzing complex education datasets and developing dashboards to support data-driven decision-making for special populations programs serving students across Texas public schools.

Graduate Research & Teaching Assistant

University of North Texas · Denton, TX

• Designed AI agents using Monte Carlo Tree Search for automated game testing — improved win rate from ~48% to ~77% through policy and rollout tuning.
• Built simulation pipelines generating 10K+ game scenarios for deck balance and strategy evaluation; research published in JBAM and adopted by Guildhouse Games.
• Mentored 100+ students on statistical and ML concepts; contributed to course redesign reducing withdrawal rates by ~35%.

Data Scientist

Lloyds Banking Group · India

• Led GBM-based fraud model under extreme class imbalance (~0.2% fraud rate) — achieved OOT AUC ~0.88, capturing ~40% of confirmed fraud in top 5% risk scores with 3+ vintage validation.
• Deployed REST API churn scoring on Azure ML, integrating with CRM to trigger retention workflows for high-risk customers identified above optimized probability threshold.
• Led end-to-end A/B test across 3.7M visits (iOS & Android) — +3% CTR lift and +4% product page visits at 95% confidence.
• Built NLP similarity and clustering system for transaction data, enabling efficient fraud pattern detection and improved reviewer experience.

Analyst, Clinical Data Systems

GlaxoSmithKline · India

• Built patient stratification models using Random Forest and XGBoost — 87% classification accuracy on clinical trial data with automated preprocessing reducing training latency by 47%.
• Designed XGBoost regression to quantify biomarker reduction in confirmed responders (MAE ~10%, R² ~0.78), supporting dose-response analysis for clinical team.
• Engineered 50+ clinical features from disparate sources; built automated monitoring framework tracking PSI, feature drift, and recall degradation monthly.

Data Engineer

UnitedHealth Group · India

• Designed scalable ETL pipelines processing 10M+ healthcare claim records daily on AWS S3 and Talend — increased processing capacity by 30% via optimized partitioning and parallel execution.
• Implemented CI/CD pipelines using Jenkins and Kubernetes for containerized data workflows — reduced deployment time by 40% and improved release reliability.
• Engineered EDI-based claims ingestion (837/835) with validation, schema standardization, and deduplication — reduced processing errors by 30%.

ETL Developer

Telkomsel · India

• Built large-scale batch ETL pipelines using IBM InfoSphere DataStage — ingested and processed 5–10M+ telecom CDR records daily for customer usage analysis and churn modeling.
• Performed performance tuning across distributed ETL jobs (parallel processing, partitioning, join optimization) — reduced data latency by ~25% within 4 months.
• Developed error-handling and recovery mechanisms with job restartability and validation checkpoints — reduced pipeline failures by ~30%.