HARSHIT GUPTA

Designed and deployed multi-agent workflows in production using Azure AI Search and Microsoft Foundry to power document intelligence and analysis across 5+ client use cases. Developed internal AI application features using FastAPI and React.js to deliver workflow APIs and user-facing interfaces for EHR/billing operations, supporting 3+ internal teams. Engineered MCP-enabled tool-calling agent with Pydantic-based structured outputs to enforce validation guardrails, capturing 8,000+ daily audit/compliance decision logs in Grafana. Enhanced Revenue Cycle Management (RCM) RPA automation pipelines using Python, Selenium, and Azure, reducing manual billing effort by 40% and cutting claim processing time by ~30% while ensuring HIPAA compliance. Benchmarking GenAI stacks (Claude, OpenAI, Gemini, vector databases) to compare latency, response quality, and framework choices for production AI workflows.

Improved claims-processing decision accuracy by 10% by integrating TF-IDF, NER, and BERT-based NLP models into production analytics workflows. Drove 15% revenue increase using 10+ interactive Power BI dashboards and ad-hoc KPI reports to provide actionable insights. Built and optimized ETL pipelines using Python and SSIS, processing 200GB+ of policy data for insurance analytics, reducing data retrieval time by 40%.

Developed academic support web platform with recommendations and GenAI Q&A services, used by 48K+ students. Increased retrieval accuracy by 55% by engineering LangChain-based RAG pipelines with hybrid retrieval, semantic embeddings (OpenAI, Cohere), and Qdrant vector search. Reduced hallucinations by 30% and improved response reliability through evaluation/observability pipelines using DeepEval, LangTrace, and Guardrails. Accomplished 30% increase in user satisfaction by building ML pipelines with PyTorch neural networks for personalized recommendations based on 200K+ student activity logs.