●

Engineered new behavioral features to improve user similarity modeling; validated performance gains via Airflow-integrated evaluation workflows.

●

Maintained and debugged Scala-based training and serving pipelines for machine learning models in a real-time bidding environment.

●

Developed RAG POC for automatic error analysis using GPT-4o, managed by LangChain, triggered by Cloud Monitoring alert via Pub/Sub, retrieving logs and indexed codebase with OpenAI embeddings and BigQuery, deployed with Cloud Run functions.

●

Built Go plugin for Kong to log timestamped API usage, caching in Redis Sorted Set, aggregating 1M+ entries/day in BigQuery.

●

Containerized applications with Docker. Deployed with Kubernetes (GKE) using Helm, 50%+ saving from VM architecture.

●

Automated CI/CD pipeline with CircleCI, managing image with JFrog Artifactory and secret with Vault, aiding agile development.

●

Designed a queueing theory-based system to forecast passenger check-in wait time and facilitate terminal capacity management.

●

Implemented multiprocessing to speed up the integration of the KDE curve 4x during queue length estimation.

●

Developed and deployed backend algorithms using Python/MySQL, achieving real-time estimation every 10 minutes and overnight forecast of backlog status, covering 194 counters, 12,000+ outgoing flights, and 165,000+ passengers per month.

●

Reviewed code, maintained Confluence document, and conducted testing of MARS (chipset inventory forecasting using GLPK), fixing bugs in linear programming constraints, and streamlining constraint matrix to reduce solution time by 40%.

●

Used Python/Cvxpy/GLPK to develop Surface Mount Technology Optimization system demo (PCBA assembly planning).

●

Implemented genetic algorithm using Python/Geatpy for Advanced Planning System demo. Modeled production scheduling problem into vehicle routing problem, solved by multi-objective mixed integer programming.