Pranav Puranik

About

Built a production LLM/RAG assistant for insurance/health claims agents that ingests 100–200 page patient PDFs via OCR (migrated from local Tesseract to Azure Document Intelligence) and delivers grounded claim detail retrieval plus summaries with PII/PHI guardrails. Experienced orchestrating large workflows with Celery worker pipelines and AWS Step Functions (S3-triggered, Fargate-based batch inference/accuracy aggregation), and collaborates closely with non-technical SMEs (claims agents/nurses) through shadowing, iterative demos, and SME-defined evaluation.

Key Strengths

• Built and deployed a production claims-agent LLM assistant for long medical/claims PDFs (OCR → chunking → retrieval → summaries)
• Designed HIPAA-conscious guardrails (PII/PHI removal, disclaimers, grounding to patient-specific data) to reduce hallucinations
• Pragmatic retrieval strategy for hard-to-embed medical codes (hybrid search + ICD lookup approach)
• Scaled document processing with async worker pipeline (Celery) and cloud OCR (Azure Document Intelligence)
• Strong orchestration experience across AWS Step Functions/S3/Fargate and Dagster for data pipelines
• SME-driven product iteration and evaluation (shadowing users, iterative demos, SME-authored test cases, relevance/retrieval metrics)
• Applied structured output validation (Pydantic) and log-driven prompt iteration

Experience

Education

Awards

• First prize - Health Care Service Corporation (Blue Cross Blue Shield) internal awards
• Recognized by VP for good performance - Health Care Service Corporation (Blue Cross Blue Shield)
• 1st prize - Mumbai open-source hackathon
• Best paper award - CS department (air-writing project)
• Best project award - CS department (air-writing project)

Skills

Languages

Publications

4 publications

Memberships

• IEEE