Umang Sharma

I’m an AI/ML-focused Data Analyst with 1.6+ years of industry experience designing and deploying NLP, Generative AI, and deep learning solutions in production. I build systems that don’t just work—they measurably improve accuracy and reliability, especially in healthcare and enterprise settings.

At Elucidata, I built a multi-agent Graph Data Science framework on the Polly Knowledge Graph, routing natural-language questions through specialized ReAct agents and orchestrating them with LangGraph tool-calling across custom GDS algorithms. Using GPT-based planner/supervisor flows with CoT prompt chaining, I achieved >80% TPR.

I also engineered an agentic pipeline for OMOP CDM 5.4 schema mapping of clinical EHR data, reaching 90%+ mapping accuracy and 85% concordance with ground truth across Synthea and MIMIC-IV. To keep the workflow compliant and dependable, I added a weighted penalty model for iterative validation, automated PHI/PII detection for HIPAA-aligned processing, and deployed the solution end-to-end as a FastAPI microservice.

Before that, I delivered a Knowledge Graph MVP using LLMs and LangChain to extract biomedical entities and relationships from PubMed/PMC abstracts, integrated into the Polly platform via an R Shiny interface. I’m equally hands-on with production RAG—building hybrid retrieval (FAISS + BM25), corrective grounding checks, and async ingestion pipelines using Celery, Redis, and Docker—while staying grounded in performance evaluation.