My FACT++ research combines hand-keypoint information at multiple stages to improve action recognition from egocentric video. It achieved 88.6% accuracy on the GTEA benchmark and was presented at IEEE ICMI 2025.
For EcoDrive, I trained PPO and SAC reinforcement-learning agents in the CARLA simulator to navigate an autonomous vehicle through waste-collection tasks. The system used an NVIDIA Jetson Nano for on-board compute.
I also fine-tuned a Vision Transformer for video action recognition and implemented cGAN, CycleGAN, and VAE models for image translation and synthetic signature generation.
At FAST-NUCES, I worked as a Teaching Assistant in Cloud Computing and as an AI Lab Demonstrator. I ran tutorials and lab sessions, evaluated assignments, and helped students with experiments and debugging.
