Muhammad A.

I build robots that navigate on their own. Day to day that means ROS2 and Nav2 autonomy stacks, and computer vision and LLM models running on NVIDIA Jetson hardware at the edge. On one robotics contract I built an autonomous object-tracking and navigation system as four Docker services: stereo depth estimation, text-prompted object detection with NanoOWL, 3D goal-pose generation, and path planning with Nav2 on top of NVSLAM and NVBLOX mapping. You could tell it what to look for in plain English and it would drive there. It ran on both x86 and ARM64, and I tested it in Isaac Sim before it touched real hardware. That client's words: "Affan helped us set up the team, and successfully brought us to the desired state in our project." The rest of my work is getting models onto hardware that was never built to run them. Speech, vision and language models on Jetson class boards under real memory limits, where the hard part is rarely the model and almost always the deployment. Where I'm strongest: - ROS2 Humble, Nav2, SLAM, sensor fusion, robot bring-up - NVIDIA Jetson Orin Nano and AGX, TensorRT, DeepStream, Isaac Sim - Object detection and tracking - YOLOv8, Grounding DINO, SAM2, OWL-ViT - On-device LLM and VLM pipelines - Whisper, Qwen, LangGraph, RAG - Docker, CI, FastAPI - the plumbing that stops a demo from staying a demo On Upwork that adds up to 100% Job Success, Top Rated Plus, 1,877 hours across 21 contracts and $60K+ earned. Several clients have brought me back for a second contract. I'm most useful when the hard part is deployment: the model works in a notebook and falls over on the target board, or the robot maps a room fine but won't plan a path through it. If that's where you are, tell me what hardware you're on and what the robot needs to do. I'll tell you honestly whether that's a week or a quarter.