• Led RL algorithm design and development for 16 out of 22 interactive scene types, driving the team to top position among competitors from MIT and UCB in the MCS competition, under the mentorship of Dr. Alan Paul Fern.
• Achieved 20 % accuracy boost to ML detectrons for human and soccer ball identification in the DARPA environment
• Implemented an Airflow scheduler to execute microservices for Unity agents, handle data preparation with Ray, and manage the ML training workflow, ensuring smooth versioning and operation within MLOps framework.