Sparsh Lamba – DataScientist at UnitedHealth group

About Me

Results-driven Data Scientist with 4 years of experience delivering high-impact AI and machine learning solutions within large-scale
enterprise environments, particularly in healthcare analytics. Proven track record of b…

Experience

Data Scientist

UnitedHealth Group (Optum) · Noida, India

Feb 2024 - Present · 2 years 4 months

BioMistral – Medical Document Classification$3M+ annual cost savings delivered by building 22 HCC classification models that replaced external vendor dependency50,000 documents daily (1M+ pages) processed through fine-tuned BioMistral-7B using QLoRA with 4-bit quantisation8% precision improvement (52% → 60%) across 40 Hierarchical Condition Category groups, outperforming legacy models5x inference acceleration achieved through TinyBERT knowledge distillation, reducing processing from 2.5 hours to 30 minutes80% GPU cost reduction by migrating from A100 to T4 hardware while maintaining model performanceClaimBot – RAG-Based Claims Intelligence AssistantBuilt from scratch a production-grade RAG system now serving 250 daily users across claims operations700+ weekly queries handled with 67% user satisfaction rate, reducing SME dependency for claims adjudication2,200 pages indexed using hybrid semantic + keyword retrieval via Azure AI SearchOpenAI reasoning models integrated to deliver structured, explainable responses for complex insurance queriesMSK Affordability – Predictive Health Deterioration$30M projected annual savings enabled through predictive model identifying high-risk patients for proactive interventions5 million patient records (232GB, 2021–2023) analysed to identify cost drivers for hip, knee, and shoulder replacements11% precision improvement (17% → 28%) in health deterioration prediction for 50,000+ high-risk members

Associate Data Scientist

UnitedHealth Group (Optum) · Noida, India

Apr 2022 - Feb 2024 · 1 year 9 months

VIVO - Scalable Call Analytics & Agent Performance Platform Objective: Build a scalable system to evaluate call center agent performance and improve operational efficiency.Played a key role in scaling the platform from 20K to 1M+ calls per day (50x growth) within a short span of 3 months.Designed and implemented machine learning models to evaluate agent performance, replacing heuristic-based systems with morescalable solutions.Developed a Call Closing Effectiveness algorithm to quantify agent behavior and improve quality monitoring processes.Engineered fault-tolerant ETL pipelines (7+) using Apache Airflow, ensuring seamless data flow across multiple business units.Leveraged Kubernetes (horizontal & vertical pod autoscaling) to optimize system performance, achieving 200-300% improvement inthroughput.Built monitoring dashboards using Grafana, enabling real-time tracking of system health and performance.Worked closely with product managers, engineering teams, and business stakeholders to continuously refine system capabilities.Business Impact:Enhanced call center productivity and decision-makingEnabled scalable performance evaluation across 400+ agentsImproved system reliability and reduced processing bottlenecks

Skills

Python Structured Query Language (SQL) Statistics Azure NumPy PySpark Transformers CI/CD Deep Learning Docker Kubernetes Machine Learning Pandas Matplotlib Scikit-Learn EDA PowerBI DOMO PyTorch NLP Generative AI LLM Fine-Tuning Knowledge Distillation Quantization RAG Architectures Agentic AI LangChain LangGraph Vector Database Azure AI Search Databricks Apache Airflow Predictive Modeling XGBoost Gradient Boosting Hyperparameter Tuning Model Evaluation Model Quantization Vector Databases A/B Testing FastAPI Power BI Data Storytelling QLoRA TinyBERT Grafana