Khang Nguyen

Designed and implemented a time-series anomaly detection system for vessel equipment using MLflow for automated model retraining; validated outputs led to significant client cost savings and helped secure additional departmental contracts. Leveraged Databricks to process and analyze large-scale datasets, developing, testing, and orchestrating over 30 automated daily data workflows to improve operational efficiency. Engineered predictive maintenance models to forecast equipment failures and assess cost impacts; conducted comparative fuel consumption analyses to highlight engine performance improvements. Developed a hybrid multiclass classification algorithm by integrating expert-defined rules with alarm signal data, combining time-series analytics and NLP to increase the accuracy and speed of alarm identification. Applied large language models to classify service reports, detect safety issues and non-conformities, and developed an RAG component to improve information extraction and enha

Lead the end-to-end delivery of the company’s first AI-driven product, collaborating with Product, UX, Data, and SRE teams to translate unstructured business challenges into a well-defined AI engineering project, achieving a public demo within 3 months. Develop a modular evaluation framework that benchmarks every layer of the RAG stack (LLM, embeddings, vector index, prompt template, retrieval strategy) to surface the best-performing combinations. Transform a naïve RAG into an agentic RAG pipeline with evaluation-first multi-retrieval and self-reflection steps, emphasizing continuous improvement and practical deployment. Deploy and scale the service by containerizing with Docker, hosting on Azure App Service, automating CI/CD via GitHub Actions, and optimizing async I/O along with effective CPU/GPU and memory management. Maintain and enhance an internal employee chatbot with 500+ daily sessions, integrating advanced features such as agent-based planning and self-evaluation while sus

Designed and implemented a time-series anomaly detection system for vessel equipment, using MLflow for automated model retraining. Processed large-scale datasets with Databricks, orchestrating over 30 automated daily data workflows. Built predictive maintenance models and conducted cost analyses to highlight improvements. Implemented a hybrid multiclass classification combining expert rules with alarm signals, and applied large language models to classify service reports, detect safety issues, and enhance information extraction with a RAG component. Collaborated with the Propulsion Digital Twin team to deliver customer dashboards and built real-time Grafana dashboards supported by Azure Data Explorer and Azure Data Factory for scalable observability.