Data Eng

Own data foundation and AI-serving infrastructure for the Healthcare Map, a national-scale intelligence platform covering 1T+ linked records, 330M+ de-identified patient journeys, and 15M+ daily clinical encounters across life sciences, payer, provider, and RWE/HEOR use cases Run Spark/PySpark and Airflow pipelines across 60+ refreshed sources: claims, EHR, lab, genomics, SDOH, mortality, and specialty feeds handling normalization, tokenization, deduplication, conflict resolution, and certification before data reaches any customer-facing layer Built publish-blocking quality gates with Great Expectations, lineage checks, and validation suites, so bad or incomplete data is caught before it reaches dashboards, APIs, or AI systems Designed RAG pipelines over certified healthcare data, combining structured retrieval from governed Snowflake marts with vector search (OpenSearch Serverless, Databricks Vector Search) over clinical concepts, code sets, methodology docs, and extracted source documents Built LLM-serving infrastructure on Amazon Bedrock and GPU-backed vLLM (SageMaker), with routing, fallback, caching, rate limiting, token budgeting, and tenant isolation, currently serving 1M+ inference requests/day at p95 2s Added PHI/PII redaction, entitlement-aware retrieval, grounded-answer checks, output validation, refusal paths, and retrieval-quality metrics to every AI-serving path Implemented offline and online feature pipelines (Spark, Snowflake, Feast-style SageMaker patterns) for real-time scoring, anomaly detection, prescriber targeting, and line-of-therapy prediction using point-in-time feature logic to prevent label leakage from late-arriving adjudicated claims; reconciled online/offline values through idempotent recomputation and backfills Partnered with data science, product, compliance, and analytics teams to publish certified data as reusable APIs, Delta Shares, and customer-facing data products, reducing insight generation from months to hours