Kevin Liu

Led development of HIPAA-compliant LLM systems for mental health triage and clinical summarization, implementing RAG and multimodal pipelines that reduced support resolution time by 32% and delivered realtime drift monitoring.

Supported Claude product experimentation and built RAG benchmarks and agentic workflows, delivering observability for latency/GPU utilization and leading robustness/red-team prompt testing for enterprise integrations.

Designed and deployed production ML systems on AWS, building distributed training pipelines for 100M+ records, SageMaker end-to-end pipelines, and prototype LLM serving stacks to improve real-time fraud scoring and CV PoCs.

Developed NLP and OCR pipelines and supported GPU-based PyTorch training for applied research projects, enabling reproducible experiments and experiment tracking for academic ML work.

Built fraud detection and real-time scoring systems for Stripe Radar, improving model precision by 22% and delivering sub-100ms scoring services with ETL pipelines using Airflow and Spark.

Prototyped and evaluated NLP classifiers for medical document classification, building scikit-learn models and annotation tooling to support biomedical text labeling.

Completed a Master of Business Administration with a concentration in finance while conducting graduate research and applied machine learning projects.

Completed a dual-focus undergraduate program combining mathematics and computer science emphasizing theoretical foundations and practical software and algorithm development.