Samuel Chen

Served as a core AI/ML engineer for Digital Workforce products, partnering cross-functionally with product, platform, security, support, and operations teams to shape requirements, architecture, delivery plans, and rollout decisions. Built an internal AI assistant for employee support and recommendations to reduce repetitive triage work by ~20%. Developed agent-based automation for multi-step support tasks by connecting LLM reasoning with internal tools/APIs, approval checkpoints, and audit logs. Designed MLOps monitoring signals, feedback capture, and rollback planning across four AI feature releases, and introduced structured settings, latency controls, and user acceptance guardrails to reduce subjectivity in rollout decisions. Prototyped model-serving and inference patterns aligned with OpenShift AI and LLM-based serving, focusing on scalable tradeoffs, request routing, observability, reliability, and practical cost/performance considerations. Reviewed AI product design/code to impr

Built NLP models for medical conversation transcription and note generation to support clinician clinical summarization, assessment-and-plan drafting, and structured documentation. Developed clinical information extraction models for medications, symptoms, diagnoses, procedures, and ICD-10 concepts, improving structured data quality for EHR work. Fine-tuned transformer-based NLP models with de-identification. Partnered with data engineering to build preprocessing pipelines for clinical text, including normalization, segmentation, speaker handling, and PHI de-identification. Designed evaluation work covering accuracy, completeness, hallucination risk, and section-level quality before broader product rollout. Implemented product analytics dashboards in Looker and Tableau to track note quality, clinician adoption, turnaround time, review patterns, and model performance trends for product and executive stakeholders.

Built NLP models for medical conversation transcription and note generation to support clinical summarization, assessment-and-plan drafting, and structured documentation. Developed clinical information extraction models for medications, symptoms, diagnoses, procedures, and ICD-10 concepts to improve quality of structured data available for EHR work. Fine-tuned transformer-based NLP models on de-identification. Partnered with data engineering to build preprocessing pipelines for clinical text, including normalization, segmentation, speaker handling, and PHI de-identification. Designed evaluation work covering accuracy, completeness, hallucination risk, and section-level quality before broader product rollout. Implemented product analytics dashboards in Looker and Tableau to track note quality, clinician adoption, turnaround time, review patterns, and model performance trends for product and executive stakeholders.