Arjun Sharma

About

ML/NLP engineer with experience at Flatiron Health building a production NLP platform that processed millions of clinical notes, using BERT/BiLSTM-CRF and spaCy to extract and normalize entities from noisy EMR text with oncologist-in-the-loop validation. Also built scalable retail ML workflows (Spark + Kubernetes + feature store caching) and applied vector databases plus contrastive-learning fine-tuning to improve retrieval relevance and recommendations.

Experience

Education

Key Strengths

• Built NLP platform processing millions of clinical notes (EMR) to structure unstructured data
• Deep clinical NLP expertise: entity extraction + normalization using BERT, BiLSTM-CRF, spaCy
• Handles noisy/inconsistent text via domain tokenization, contextual embeddings, and iterative expert feedback
• Designs entity-resolution pipelines combining rules, fuzzy matching, and embedding similarity
• Strong evaluation discipline: precision/recall/F1 on held-out sets plus domain-expert review
• Improves retrieval/linking with vector DBs and contrastive-learning fine-tuned embeddings
• Production-grade ML/data workflows with orchestration (Airflow/Kubeflow), testing, monitoring, versioning
• Proven scalability optimization: Spark + Kubernetes + feature-store caching reduced runtime from hours to minutes
• Clear prototype-to-production criteria balancing metrics, validation, latency, and operational risk