Svachuta Gollavilli

About

Built a production LLM-powered RAG agent for healthcare/insurance operations that retrieves and summarizes patient medical documents with grounded citations, scaling to ~4.5M records. Addressed medical shorthand and terminology by fine-tuning ~120 lightweight DistilBERT models by specialty and validating entities against SNOMED/RxNorm, while using SHAP/LIME and human-in-the-loop review to make decisions explainable to stakeholders.

Experience

Education

Key Strengths

• Built and deployed production LLM RAG agent for medical/insurance operations
• Scaled embedding + retrieval over ~4.5M medical records with optimized chunking and vector DB design
• Domain-specific NLP approach: fine-tuned ~120 lightweight DistilBERT models by medical specialty to handle shorthand/nomenclature
• Improved decision turnaround time to under 90 seconds for repeated claim/treatment queries
• Strong experimentation/POC-driven iteration across RAG strategies (splitting, indexing, retrieval) to improve accuracy and consistency
• Explained model decisions to non-technical stakeholders using SHAP/LIME and implemented human-in-the-loop oversight