Sidhant Thakur

Led end-to-end data science initiatives including an XGBoost-based fraud detection model that automated processing of 10K+ claims per day and reduced manual reviews by 60 hours/month. Designed a Kafka-to-Snowflake ingestion pipeline handling 1.5M+ transactions daily for real-time drug-efficacy monitoring. Built a PyTorch anomaly-detection model to flag supply-chain inconsistencies across 20+ sites within 15 minutes, cutting incident-response time by 30%. Implemented TensorFlow time-series models forecasting demand for 100+ pharmaceutical products to optimize inventory. Delivered real-time clinical and claims insights via Power BI dashboards, and developed a SQL/Python predictive model that increased high-risk patient identification by 20%. Added LLM-based summaries to shorten document review by 45% and a lightweight embedding-based semantic search to speed record retrieval by 40%. Created a RAG-style evidence-retrieval workflow that reduced evidence gathering time by 45% and improved d

Drove a Scikit-Learn-based fraud detection system on Python/Hadoop handling 10M claims/day with ~5-second anomaly flags, delivering substantial cost savings. Built a PyTorch-based deep-learning classifier on GCP BigQuery processing 500K+ medical records/month to improve claim approvals. Automated Spark ETL pipelines, reducing transformation time from 6 hours to 30 minutes. Created a large-scale risk-stratification model using AWS EMR and Tableau to analyze 1B+ patient interactions for data-driven coverage decisions. Implemented K-Means segmentation to identify 5 cohorts for targeted interventions, and maintained a Tableau dashboard suite with 25+ dashboards for rapid decision-making. Developed regression-based predictive analytics improving forecast accuracy by 22% and reducing fraudulent payouts by 18%. Deployed LLM-based summarization to cut document review time by 60%, and built a lightweight embedding-based search tool to halve lookup times. Introduced an LLM-powered Q&A workflow t

Developed Scikit-Learn models to predict failures across 100+ product lines, reducing warranty claims by 15%. Led pricing analysis on 3M+ transactions using R and ggplot2 to refine global pricing strategy and margins. Implemented a Random Forest classifier for customer segmentation to drive targeted campaigns. Designed an Azure-based data lake to centralize data and accelerate ML pipelines, cutting decision speed by 30%. Migrated 250+ Excel reports to PostgreSQL with automated pipelines for real-time reporting, reducing reporting time by 60%. Performed anomaly detection on hardware logs using Hadoop/Spark to uncover defect patterns and improve maintenance planning. Enhanced Excel automation with macros to deliver predictive sales insights, reducing manual effort by 70%.