Robotics · Perception · Embedded AI
Hi, I'm Ismail
I am a computer vision engineer building real-world perception systems in C++ and Python. My work spans geometric vision, multi-camera bird's-eye-view stitching, calibration, robotics, and embedded inference on NVIDIA Jetson platforms.
I care about systems that are simple to reason about, fast to deploy, and accurate enough to survive real field conditions.
01 / Work
Oreyeon — Computer Vision Engineer
- Redesigned and optimized a C++ multi-camera image stitching library for large-scale BEV generation, achieving approximately 98% speedup.
- Used algorithmic optimization, multithreading, and GPU acceleration to support image sequences exceeding 3,000 frames without drift.
- Optimized and deployed PyTorch models in C++ using TensorRT 8 on embedded NVIDIA Jetson platforms, reducing end-to-end runtime by about 35%.
- Redesigned calibration and refinement protocols to improve geometric consistency and long-term calibration stability.
- Evaluated IMU- and LiDAR-based approaches for camera pitch compensation, and built a vanishing-point fallback when inertial data was unavailable.
- Developed a measurement pipeline combining BEV stitching, RTK-GPS fusion, and pinhole-based geometric modeling, reaching error below 6%.
- Built Python and C++ validation tools for accuracy checks, regression testing, benchmarking, and offline dataset analysis.
- Worked on secure deployment readiness, including STIG-related system hardening and configuration.
- Led a month-long on-site product validation campaign in Portugal under real operating conditions.
Course Content Writer
Created technical and educational course content before joining Oreyeon full-time.
Tutor
Tutored mathematics, physics, chemistry, and engineering subjects. This helped me build the habit of explaining technical ideas clearly and from first principles.
Rafik Hariri University — Laboratory Assistant
Supported university laboratory work and technical instruction in an academic engineering setting.
ISS Software Hive — Internship Trainee
- Built PyTorch models including CNNs, RNNs, and GANs.
- Implemented reinforcement learning algorithms including Q-Learning, DQN, and PPO.
- Trained Unity agents using ML-Agents and custom reinforcement learning implementations.
02 / Selected projects
Humanoid Robot: Vision-Based Pick-and-Place
Built an end-to-end pick-and-place system for a simulated humanoid robot in MuJoCo. Replaced direct access to simulator geometry with an RGB-D perception stack using color segmentation, depth back-projection, and 3D target fusion to locate objects and the destination table.
Combined PPO-based navigation with walking, crouching, and reaching controllers, coordinating base positioning, grasping, object transport, and drop-off through a staged task controller.
OpenCV Robust 2D Translation Estimator
Contributed cv::estimateTranslation2D() to OpenCV's calib3d
module, implementing RANSAC and LMedS robust translation estimation with unit tests,
performance benchmarks, and Python bindings. Pull request:
PR #27950.
Stereo Visual Odometry System
Built a stereo visual odometry pipeline in C++ using OpenCV, Eigen, and PCL, implementing feature matching, relative pose estimation, triangulation, and 3D trajectory visualization.
General Pick-and-Place Mobile Robot
Developing a mobile robot capable of autonomous navigation and pick-and-place tasks using ROS and computer vision. Leading software development.
SLAM Mobile Robot Using ROS and Gazebo
Simulated a two-wheeled robot in Gazebo for mapping with Gmapping, localization with AMCL, and navigation using move_base.
Reinforcement Learning Experiments
Implemented and trained agents using Q-Learning, DQN, PPO, OpenAI-style environments, Unity ML-Agents, and custom reinforcement learning code.
Embedded Deep Learning Deployment
Converted and validated PyTorch model behavior in C++ deployment pipelines using custom preprocessing, postprocessing, TensorRT optimization, and NVIDIA Jetson hardware.
Multi-Camera BEV Perception
Worked on multi-camera image stitching, calibration refinement, geometric alignment, camera pitch compensation, and long-sequence validation for bird's-eye-view perception.
03 / Education
American University of Beirut (AUB)
Master of Engineering in Artificial Intelligence and Robotics.
Rafik Hariri University
ABET Accredited Bachelor of Engineering in Mechatronics Engineering. Final grade: 92.37/100.
Beirut Annunciation Orthodox College
Lebanese Baccalaureate.
Hariri High School II
Secondary education.
04 / Technical skills
- C++
- Python
- Linux
- Git and GitHub
- OpenCV
- CUDA
- TensorRT 8
- PyTorch
- ROS
- Gazebo
- MuJoCo
- Unity ML-Agents
- Eigen
- PCL
- MATLAB
- SolidWorks
- AutoCAD
- Fusion 360
- Arduino
- Doxygen
05 / Certifications
- MATLAB Onramp — MathWorks, June 2022
- Intro to Machine Learning — Kaggle, May 2022
- Code Foundation for ROS Path — The Construct, July 2021
06 / Interests
- Geometric computer vision
- Robotics perception
- SLAM and visual odometry
- Embedded AI deployment
- Fast C++ vision systems
- Multi-camera perception
07 / Contact
For work, research, or project discussions, get in touch.