Phani K

About

Built and deployed a production LLM-powered clinical insights/summarization assistant for healthcare teams, including a Spark+Airflow pipeline, fine-tuned transformer models, and a FastAPI Docker service on AWS. Demonstrates strong MLOps/LLMOps depth (Airflow on Kubernetes, custom AWS operators/IAM, MLflow, CloudWatch) and practical reliability work like hallucination mitigation, confidence scoring, and retrieval-backed evaluation with shadow deployments.

Experience

Education

Key Strengths

• Built and deployed an LLM-powered clinical insights assistant to production on AWS
• End-to-end ML/LLM pipeline design (Spark + Airflow) for unstructured clinical notes
• Production inference optimization via distillation and quantization to reduce latency
• Hallucination reduction using rule-based validation, confidence thresholds/scoring, and improved retrieval
• Airflow orchestration at scale: optimized long-running DAGs with task parallelism and Kubernetes executors
• Deep AWS integration for ML workflows (S3, SageMaker, ECS) including custom Airflow operators and IAM role design
• Strong monitoring/observability practices (CloudWatch + Airflow alerting) for ingestion and training pipelines
• Structured evaluation approach: unit tests, offline eval, scenario tests, A/B and shadow deployments with HITL review
• Effective collaboration with non-technical clinical stakeholders using prototypes, visuals, and expectation-setting