Sneha Vandanaa Vijayakumar

Led the end-to-end research and development of an AI system designed to address the critical healthcare challenge of predicting 30-day hospital readmissions. The goal was to surmount the limitations of traditional models by integrating unstructured clinical context with structured data to improve patient safety outcomes. Key Responsibilities & Architecture: • Designed a “Cohort-Specific Multimodal Fusion Framework” to process diverse data streams from the MIMIC-III dataset, including 48+ structured EHR features and unstructured discharge summaries. • Built a Retrieval-Augmented Generation (RAG) engine to process clinical narratives using GCP with GPU acceleration, ClinicalBERT for dense vector embeddings, and FAISS for semantic indexing. • Integrated Mistral-7B via Ollama for clinical reasoning and developed a dynamic routing mechanism to inject cohort-specific guidelines into prompt pipelines. • Developed distinct “specialist” ML models (Random Forest, LightGBM, CatBoost, etc) for high-risk cohorts and engineered a late-fusion weighted ensemble to combine their probabilistic outputs with the RAG signals. Outcomes & Impact: • Successfully optimized the system for a safety-first clinical setting, achieving a 92.1% Recall and 90.7% ROC-AUC. • The fusion model successfully identified 492 critical high-risk cases that were missed by standard unimodal baseline models. • Enhanced clinical trust by integrating SHAP (SHapley Additive exPlanations) values to provide transparent, feature-level explanations for model predictions.

As an IT Data Analytics Intern, I was responsible for transforming and modernizing the company's product data quality reporting. I designed and built a robust reporting framework on modern EDH (Redshift) and Tableau platforms, directly converting six critical legacy reports. This project had a significant impact: - Automation: I rewrote existing SQL routines to automate daily data updates and retain a historical log, eliminating the prior manual effort of 1-3 days per report, which was previously a quarterly process. - Data Visualization: I developed interactive Tableau dashboards that transformed static CSVs into dynamic, navigable reports. These dashboards featured a "wall of numbers" view and drill-down capabilities, enabling a detailed analysis of data quality errors. - Efficiency: By optimizing the database design within the new framework, I reduced the time needed to develop additional reports by 50%, ensuring a scalable solution for future reporting needs. - Cross-Functional Collaboration: I conducted a comprehensive SWOT analysis for a CFO case study on a key product, collaborating with interns from Finance and Supply Chain. This experience required me to synthesize diverse perspectives into a cohesive, comprehensive presentation, showcasing my ability to work effectively across business functions. My work enhanced data integrity and fostered a foundation for future AI/ML initiatives by ensuring a higher quality of data within the Oracle R12 system. This internship provided valuable hands-on experience in data analytics, business intelligence, and collaborative problem-solving within a corporate environment.