Amrutha Vinayakam

Pennsauken, NJ

• Architected a centralized Azure data warehouse, designing scalable schemas, ingestion processes, and standardized reporting layers, reducing data discrepancies by 15% and saving 8 hours/week.
• Built an ensemble of Random Forest and Logistic Regression models to segment agencies on demographic and behavioral data, driving a 15% increase in distribution across 200+ partner agencies.
• Deployed an XGBoost demand forecasting model on historical distribution and program data, improving planning accuracy and directing resource allocation decisions for weekly operational reviews.
• Delivered executive-ready Power BI dashboards and governed KPI reporting, enabling weekly and monthly operational reviews by translating program needs into reusable analytics views.
• Aligned Grants, Marketing, and Finance stakeholders on shared KPIs, converting data insights into program strategies adopted in leadership decision-making across the organization.

Florida, Remote

• Elevated user behavior prediction by 25% by developing an ensemble of XGBoost and Logistic Regression models on user, property, and policy data, with LLM-based feature enrichment for program rules, while optimizing for latency and cost.
• Increased match accuracy by 10% by segmenting users with K-Means clustering for program alignment, validated with t-tests to ensure statistical significance of key insights.
• Engineered an automated evaluation suite using F1, confusion matrices, and cross-validation, ensuring model reliability across releases; maintained a 95% testing coverage rate.

New York, NY

• Developed and maintained 5+ ETL data pipelines for academic datasets, reducing data quality issues by 40% by standardizing ingestion, cleaning, and data-validation steps for 30+ students.
• Created 4+ interactive Tableau dashboards for feedback and enrollment KPIs, improving faculty decision-making efficiency by 50% by consolidating metrics into self-serve visual reporting.
• Implemented a Transformers-based classification model, achieving 92% accuracy by fine-tuning Hugging Face models to automate research content tagging and save 10+ hours/month.

New York, NY

• Accelerated model quality by 25% and real-time processing speed by 30% by iterating on feature pipelines and training workflows using Python, SQL, Scikit-learn, TensorFlow, PyTorch, and Keras.
• Boosted production chatbot response accuracy by 15% by refining NLP pipelines with Hugging Face Transformers and the OpenAI API, including prompt and retrieval tuning.
• Reduced inference latency by 20% by preprocessing and optimizing text pipelines using SpaCy and NLTK for large-scale datasets.
• Optimized pipeline efficiency by 35% by integrating LangChain and LlamaIndex components to increase throughput and support scalable agent workflows.
• Drove up user support engagement by 15% by contributing to AI virtual agent features that expanded real-time assistance coverage.

India

• Sharpened monthly revenue projections by 20% by building financial forecasting models from historical claims and billing data for chronic care programs.
• Productionized forecasting models (ARIMAX, SARIMAX) and XGBoost in Azure ML Studio for repeatable training and scoring, automating monthly runs to support revenue planning.
• Presented ROI models to cross-functional finance stakeholders, influencing $2M+ investment decisions across 3 digital health pilots.