Industry
Reinforcement Learning for Energy Optimization
Developed PPO-based reinforcement learning policies for intelligent building energy control. Worked across the full ML lifecycle, including simulation, offline policy evaluation, AWS deployment, and production monitoring.
This work focused on applying reinforcement learning to intelligent building energy control. The objective was to learn control policies that could balance energy use with operational requirements in a dynamic environment.
Policy development
I developed Proximal Policy Optimization (PPO) policies in simulation, working across state and action design, reward formulation, training, and policy evaluation. The system treated building control as a sequential decision-making problem in which present actions affect both immediate consumption and future system behavior.
Evaluation
Before deployment, candidate policies were evaluated offline against simulated and historical operating behavior. This made it possible to compare policies, inspect failure modes, and assess whether improvements remained consistent across operating conditions.
Production lifecycle
The work extended beyond model training to AWS deployment and production monitoring. The resulting workflow connected experimentation, policy validation, deployment, and ongoing observation of system behavior.