Machine Learning Engineer
CurrentML Pipeline Construction• Converted Scala-based MF model to Python-based ML model• Designed and developed Python framework structure (Python, PyTorch)• Built model training/experimentation environment on NVIDIA HGX A100 servers (Argo Workflows, JupyterHub)Inference-Dedicated Server Cluster Setup• Built inference-dedicated server cluster to separate model inference from recommendation API with business logic• Constructed TorchServe cluster on AWS EKS (CPU-only… Show more ML Pipeline Construction• Converted Scala-based MF model to Python-based ML model• Designed and developed Python framework structure (Python, PyTorch)• Built model training/experimentation environment on NVIDIA HGX A100 servers (Argo Workflows, JupyterHub)Inference-Dedicated Server Cluster Setup• Built inference-dedicated server cluster to separate model inference from recommendation API with business logic• Constructed TorchServe cluster on AWS EKS (CPU-only mode)• Changed communication data format from JSON to msgpack, improving communication speed by 30%• Packaged code to enable reuse of model training code in TorchServe inference handlersRecommendation API Development, Monitoring, and Deployment Automation• Developed recommendation API for service servers (Scala)• Created dashboard for recommendation KPI monitoring (Google Looker Studio)• Implemented API server performance tracking (Datadog)• Established and operated CI/CD for stable recommendation server operation (Jenkins - Configuration as Code, Pipeline as Code)Friend Recommendation Model Development using LightGCN• Replaced computationally expensive MF-based taste matching score friend recommendation with ML model-based friend recommendation• Enabled more diverse user recommendations due to relationship-based approachData Analysis Team Building and Leadership• Established data analysis team, set team agendas, recruitment, and management• Developed KPIs and dashboards using Google Looker Studio• Operated team to focus on data analysis for product development, content acquisition, and IR/PR needs, reporting Show less