Corpsy, Brentwood, TN
Care First, Blue cross blue sheild,, Baltimore, MD
Data Scientist: Ups Atlanta, GA
Data Scientist, Macy's, Irving, Tx
- Designed and implemented scalable end-to-end ML pipelines for clinical note extraction and summarization using TensorFlow, PyTorch, and MLflow for tracking and model management.
- Deployed transformer-based NLP models (GPT, BERT, T5) to production using Docker, Kubernetes, and RESTful APIs with Flask/FastAPI for secure, compliant access.
- Built real-time and batch inference pipelines with Apache Kafka, Spark, and Airflow, enabling clinicians to retrieve insights with low latency.
- Fine-tuned domain-specific LLMs (GPT-4, LLaMA, Falcon) on medical records and research to support summarization, Q&A, and knowledge retrieval tasks.
- Utilized HPC environments and GPU clusters to accelerate training and inference of transformer-based NLP models on large-scale clinical datasets.
- Developed healthcare-specific Agentic AI assistants capable of autonomous patient data retrieval and clinical summarization.
- Built multimodal AI solutions combining medical images, clinical notes, and structured patient data using GPT-4o and Gemini.
- Implemented HIPAA-compliant LLMOps pipelines with automated monitoring, evaluation, and governance controls.
- Developed AI-powered search platforms leveraging vector databases and semantic retrieval for clinical knowledge management.
- Built automated evaluation pipelines using RAGAS, LangSmith, and OpenAI Evals to measure answer quality and factual consistency.
- Designed distributed data and model pipelines using Spark, Dask, and Kubernetes to process and serve real-time clinical insights at scale.
- Implemented parallelized and GPU-optimized RAG workflows for clinical chatbots, reducing latency for evidence-based physician queries.
- Applied GitHub Copilot and Anthropic Claude for pair-programming and faster pipeline prototyping, accelerating platform modernization efforts.
- Applied prompt engineering and few-shot learning to adapt general LLMs for specialized medical terminology with minimal additional labeling.
- Created robust NLP pipelines for named entity recognition, topic modeling, and sentiment analysis on patient feedback using spaCy, Hugging Face, and NLTK.
- Integrated transformers like BERT, T5, and GPT-4 into applications for discharge summaries, claim validation, and medical virtual assistants.
- Built resilient data pipelines for structured and unstructured EHR data using SQL, dbt, Spark, AWS Glue, and GCP Dataflow.
- Leveraged AWS, GCP, and Azure for distributed training, auto tuning, and scalable inference, ensuring HIPAA compliance and operational efficiency.
- Deployed monitoring with Prometheus and custom dashboards to track model drift, data health, and latency for reliable clinical insights.
- Developed interactive dashboards using Streamlit, Dash, and Power BI to present outputs like entity extraction, sentiment trends, and query results.
- Integrated SHAP and LIME for model explainability, helping physicians and compliance teams validate AI-generated answers and summaries.
- Automated CI/CD workflows for ML pipelines with Airflow, MLflow, Git, and container registries to streamline deployment and rollback.
- Explored generative AI applications (e.g., text-to-image for patient education) using Stable Diffusion, DALL·E, and CLIP to support marketing and engagement.
- Created prompt-engineered LLM-based medical data summarization tools using OpenAI, Gemini, and Anthropic APIs with custom evaluation metrics via LangSmith.
- Built LangGraph-powered clinical assistants for summarization and diagnostic automation; developed dashboards using Tableau and Streamlit for risk trend analysis.
- Integrated clinical AI pipelines into enterprise systems, including ServiceNow and cloud platforms, for real-time decision support and workflow automation.
- Designed and executed prompt engineering experiments to optimize LLMs for healthcare use cases, collaborating with SMEs for clinical evaluation.
- Built an experimental framework to test and tune generative AI models, applying offline and online evaluation metrics for performance benchmarking.
Environment: GPT-4o, Claude, Gemini, Lang Graph, Llama Index, Pinecone, Chroma DB, Lang Smith, RAGAS, OpenAI Evals, Vertex AI, Azure OpenAI, AWS Bedrock, PyTorch, TensorFlow, Kubernetes, Kafka, Spark, Airflow, MLflow.
