Sumit Kumar

• Led cross-functional team to conceptualize, refine, and integrate data science methodologies, resulting in pioneering solutions that addressed intricate problems and yielded innovative outcomes
• Elevated strategic planning and executives' decision-making by scaling analytical capabilities across all business areas, quantifying average 40% enhancement in key performance indicators (KPIs)
• Collaborated with internal stakeholders to identify and gather analytical requirements for customer, product, and project needs, leading to 25% reduction in project scope changes and 15% increase in on-time project deliveries

Patient Readmission Prediction:

• Built XGBoost model to forecast 30-day readmissions for 50k patients, yielding 86% accuracy
• Integrated diverse EMR data, enhancing predictive ability
• Merged and harmonized hospital metrics, diagnoses, surgeries, and other data points to develop accurate readmission predictions within a critical timeframe, reducing readmission rates by 25% and improving patient outcomes.

Discovering Recurring Patterns in SR Incident Tickets:

• Managed 500K SR Incident tickets with NLP. Used BERT, auto-encoders, and HDBSCAN to cut manual effort by 50%.
• Employed BERT embedding and clustering techniques to effectively group SR incidents, resulting in 30% reduction in incident resolution time and enhancing customer satisfaction scores by 25%

AIOps: Enhancing Client Experience and Issue Resolution:

• Combined big data and ML for automated operational tasks. ML algorithms predicted crashes, reduced mean time to resolve by 60%.
• Utilized Isolation Forest algorithm to proactively detect anomalies in operational data, delivering timely alerts that saved almost $200,000 in potential losses due to operational disruptions.
  

Advanced Software Issue Analysis and Recommendation System:

• Developed CR recommendation using NLP and ML on 3 lakh call stack logs. Enhanced issue resolution with 10% F1-score improvement.
• Implemented NLP and ML techniques, including TF-IDF and Convolutional Multi-label Sentence Classification models, to identify and apply accurate fixes, reducing development time by 25% and stabilizing system performance.

NASH Disease Prediction:

• Engineered high-performing binary classifiers using XGBoost, logistic regression, decision trees, & random forests, achieving an 85% AUC score in determining NASH status from clinical lab data
• Discovered 30% of NASH cases in real-world data, with impressive 82% sensitivity rate.

Clinical Trial Enrollment Optimization:

• Designed and implemented optimization model utilizing Poisson Gamma Analysis, taking an average 20% reduction in trial duration and increasing likelihood of meeting enrollment goals
• Utilized model's adaptability to re-calibrate enrollment projections with historical and real-time data, enhancing decision-making and yielding notable 15% trial success rate improvement within set duration.