• Design and operate ML platforms that support end-to-end workflows: data ingestion, feature engineering, training, evaluation, deployment, and monitoring. • Build and maintain CI/CD for ML (testing, packaging, versioning, reproducibility, automated rollbacks, approvals). • Implement MLOps best practices: model registry, experiment tracking, lineage, governance, and reproducible training environments. • Develop scalable training infrastructure (distributed training, GPU scheduling, cost controls, auto-scaling). • Create and maintain feature pipelines / feature stores, ensuring consistency between training and inference (training-serving skew prevention). • Establish model monitoring and observability: performance, drift, bias/fairness signals (where relevant), latency, throughput, and data quality. • Build and own end-to-end LLM delivery pipelines: prompt/versioning, retrieval, orchestration, evaluation, deployment, monitoring, and iterative improvement. • Create robust LLM evaluation harnesses (offline + online): golden datasets, automated regression testing, human-in-the-loop review workflows, and risk scoring. • Build cost controls: token/cost budgeting, caching strategies, autoscaling, and performance tuning. • Productionize ML Models on GCP using containers and orchestration (e.g., GKE, Cloud Run), and build CI/CD for ML/LLM systems with automated tests and safe rollouts. • Implement observability: tracing, metrics, logs, dashboards, alerting for model/system health (latency, token usage, error rates, retrieval quality, hallucination indicators, drift where relevant). • Design systems with security and privacy by default: IAM, least privilege, secrets management, audit logs, encryption, data retention, and PHI/PII handling. • Implement governance: model/prompt lineage, dataset provenance, evaluation traceability, and approval workflows aligned with healthcare compliance expectations. • Integrate guardrails: content filters, policy checks, prompt injection defenses, structured output validation, and fallback strategies.