AI Engineer & Roboticist
Building intelligent systems that perceive, reason, and act in the real world.
I'm a Fulbright Scholar pursuing my Master's in Robotics at Northeastern University, with a background in mechanical engineering and 1.5 years of experience shipping production AI systems.
At Smart Forum, I led a 12-person cross-functional team to launch Clear Minds—Pakistan's first AI-powered drug prevention app. I've built multi-agent automation pipelines, NLP-powered RAG chatbots, and custom voice cloning systems optimized for real-time inference. My coursework has taken me deeper into SLAM, motion planning, and robotic control.
Outside of work, I travel when I can and have, on one memorable occasion, thrown myself out of an airplane at 14,000 feet. I'm currently looking for Summer 2026 internship opportunities where I can work on robotics or AI systems that operate in the physical world.
Sep 2022 – May 2023
Developed assistive system enabling visually impaired users to grasp objects via voice commands; trained custom object detection model using transfer learning on SSD MobileNet V2 achieving 83.2% mAP, converted model to TFLite for edge deployment, implemented contour-based rotation angle extraction for gripper alignment, and integrated with 6-DOF Niryo One arm for autonomous pick-and-place.
Designed and analyzed state-space controllers for a self-balancing Segway (inverted pendulum), progressing from PID to observer-based full-state feedback; quantified instability margins, actuator saturation effects, and power trade-offs, then validated a Luenberger observer that recovers full-state performance from tilt-only measurement.
Implemented TD3, SAC, and PPO from scratch in PyTorch on HalfCheetah, Hopper, and Walker2d (Gymnasium MuJoCo); ran multi-seed training with learning curves, sample-efficiency analysis, random-policy baseline, and hyperparameter sensitivity sweeps — no Stable-Baselines3.
End-to-end natural language to MuJoCo execution pipeline: Claude maps commands to a validated JSON motion plan over fixed primitives for a 7-DOF Franka Panda arm, with damped Jacobian IK, chained Cartesian waypoints, and stable block stacking — no ROS or learned policies
Reproduced NeurIPS 2024 Visual SLAM paper; achieved 1.23 cm trajectory error with YOLOv8 segmentation, outperforming reported 1.6 cm by 23%.
Implemented forward kinematics, trajectory generation, numerical Jacobian, and Newton's method inverse kinematics solver for 6-DOF arm; tracked 400-waypoint trajectory with sub-mm precision.
Designed single-actuator grasping mechanism with cascading four-bar linkages; derived analytical forward kinematics for closed-loop system and validated with 3D-printed prototype.
DeepLearning.AI
AWS
MIT (in-person)
I'm always open to discussing new opportunities, research collaborations, or interesting projects.
