Sep 2020 - Present
- Constructed LGBM models with direct recursive hybrid strategy to forecast sales 9 months ahead, obtaining under 15% error during pandemic periods and enhancing average 30% allocation precision.
- Built the ATP optimization tool with Linear CBC (Branch-and-Cut) Solver by the python ortools package on a scale of some 4,000 lines, speeding up weekly routines of 2-day workloads to within 1 minute.
- Ran a PoC of CI/CD on cloud platforms inclusive of AWS, GCP, and OCI, contributing to comparison report evaluating 3 platforms through time, cost, EoU, IAM, compatibility, etc.