Scope
My work covered several decision problems in e-commerce where a platform must choose among actions, observe uncertain outcomes, and adapt over time.
Selected problem families
- Dynamic pricing and seller selection.
- Advertiser or winner selection in marketplace allocation.
- Contextual bandit and reinforcement-learning formulations.
- Delivery, supplier allocation, and routing optimization.
What I learned
These projects strengthened a principle that continues to shape my work: an ML model should be judged by the decision process it supports. Reward design, constraints, delayed feedback, and operational feasibility matter as much as predictive accuracy.
Teaching connection
During the same period, I taught reinforcement learning, mathematical optimization, and mathematics for AI, using practical examples and Python implementation to connect formal methods with decisions.