Benchmarking and evaluation
I compare LLM and VLM stacks across NVIDIA, Apple Silicon, and AMD hardware, then turn the results into repeatable recipes, dashboards, and concrete engineering feedback.
04
Earlier projects
Reinforcement learning
Multi-Agent Penalty Kick
Trained striker and goalkeeper agents together with Q-learning in V-REP, exploring competing policies, discrete action spaces, and reward design.
Planning
Hierarchical Package Delivery
Decomposed vehicle routing into package clustering and delivery routing, comparing two solver pipelines implemented with ROS and simulated in Gazebo.
Computer vision
Autonomous Harvesting Robot
Built an E-Yantra robot that detected requested fruit on a Raspberry Pi, plucked it with a gripper, and delivered it to assigned arena coordinates.
Assistive robotics
Staircase-Climbing Wheelchair
Built a gyro-stabilized, Arduino-controlled prototype with a robotic arm in 24 hours; placed first among 30 national teams and later earned IEEE Best Hackathon Project.