I make robots move fast, precisely, and in the real world. Right now I'm building high-dynamic capabilities for humanoid robots at Iowa State University.
My broader mission is high-dynamic motion on humanoid robots. These are the explosive, whole-body skills still out of reach for machines built to walk carefully.
I use trajectory optimization and model-based control, so I can reason directly about the torque and timing behind each motion. Alongside the research I'm building a custom humanoid and moving into reinforcement learning.
Underneath it all, I'm studying classical control to understand how robots and controllers really work, before moving into physical AI.
I chose Mechatronics Engineering at NUST when it wasn't a conventional or well-understood path in Pakistan. I went with it anyway. It's a field that fuses hardware, firmware, AI, and software into systems that live in the real world.
The turning point wasn't a grade. It was curiosity. Once I started asking questions and building things, it led to a research opportunity, a journal paper, and a paid internship. An international internship in generative & discriminative AI for medical imaging made it clear: I didn't want abstract code. I wanted hardware, robots, and systems I could build, touch, and improve.
Six intense months as an AI Engineer at a Y Combinator startup taught me to move fast and take ownership. Then in August 2025 I moved to the United States for a PhD in Computer Science focused on robotics and AI, and I haven't looked back.
I was the only mechatronics engineer from across Pakistan selected for an on-site research internship in Canada, at the University of Calgary under Dr. Matthias Wilms in summer 2024.
There I worked on generative and discriminative AI for medical imaging: explainable retinal-disease detection at 93% accuracy, and latent Flow-Matching generative models built on a Stable Diffusion 2 VAE with the MONAI codebase. The work was accepted at SPIE Medical Imaging 2025.
It was also a turning point. Working abroad gave me deeper biomedical experience and confirmed that my future was in robotics and physical systems, not purely abstract AI.
PhD Researcher & Robotics Research Assistant · United States
PhD in Computer Science, robotics specialization, with a 4.00 GPA so far. My work centers on high-dynamic humanoid motion using model-based control, a custom humanoid build, and a growing push into reinforcement learning. I also teach as a robotics and data-science TA.
AI Engineer → Associate Product Manager · Islamabad
Built a real-time Voice AI agent end-to-end (Whisper, Kokoro TTS, turn detection, LangChain), designed GAN-based synthetic-data pipelines, and deployed a 4-bit-quantized SmolVLM2 in Docker.
On deployment I optimized YOLO with TensorRT for a 57% latency cut (17.2 → 7.4 ms) and ran multi-PC Docker Swarm clusters at sub-50 ms/request, then stepped into an APM role across product and go-to-market.
AI Research Intern · Canada
The only mechatronics engineer from Pakistan selected for this on-site internship. Built explainable retinal-disease detection (93% accuracy) and latent Flow-Matching generative models (Stable Diffusion 2 VAE + MONAI). Work accepted at SPIE Medical Imaging 2025.
Research & AI Intern · Islamabad
Built a semi-automatic image-segmentation labeler with SAM, applied Vision Transformers to solar-panel defect detection (journal paper), and delivered a collaborative Edge-AI enclosure with multi-sensor acquisition on Raspberry Pi.
Designed and fabricated a pair of robotic arms end to end, from the mechanical build to the control stack, as a platform for manipulation research.
Built a low-latency teleoperation setup for intuitive control of robotic arms, used both to demonstrate and to collect demonstrations for learning.
Brought UR10e and SO-101 manipulators into high-fidelity simulation for control and imitation-learning research before touching hardware.
Built pipelines that turn teleoperated demonstrations into learned manipulation policies, closing the data-to-policy loop for robot skills.
Designed protective structures ("armor") for humanoid robots so aggressive, high-dynamic experiments can run without wrecking the hardware.
Automated a baseball-throwing rig for repeatable, high-speed dynamic-motion experiments and data collection.
The future of robotics starts with early curiosity. I built a free, interactive course that breaks physical AI and robotics down into hands-on adventures kids can actually play with. Real on-device AI models, no accounts, no downloads.
Explore AI for Kids schoolI break down robotics tooling, simulation, and deployment for a growing community of 970+ followers. Longer, more technical write-ups live on the blog.
A robot slides and balances a tray with no grasp, and the ADMM split that turns 2¹⁴⁰ contact modes into five small problems solved in parallel.
Read on the blog → 393 reactionsA low-cost, open-source quadruped for education and research, and why open hardware matters.
Read on LinkedIn → 151 reactionsDocker for robotics: reproducible environments, sensor integration, and prototype-to-production.
Read on LinkedIn → 96 reactionsWhich simulator to reach for, weighing physics fidelity against GPU-scale training throughput.
Read on LinkedIn → 50 reactionsA curated list of 8 buildable open-source robot arms for researchers and hobbyists.
Read on LinkedIn →If robotics, humanoids, or dynamic motion is your thing, reach out. Even if a collaboration isn't possible, there's always something we can learn from each other.