Available for work
محمد المصيطف
مهندس روبوتات وذكاء اصطناعي
أبني أنظمة روبوتية ترى وتخطط وتتحرك.
- Robotics
- Artificial Intelligence
- Machine Learning
- Deep Learning
- Computer Vision
- NLP
About
I'm an engineer working at the intersection of robotics and applied AI. My work runs the full width of the problem — training the perception and language models, then earning the right to trust them by putting them on hardware that moves through the real world.
On the AI side that means computer vision (detection, segmentation, visual SLAM), deep learning (CNNs, transformers, reinforcement learning), and increasingly NLP and LLMs for grounding natural-language instructions into robot behaviour. On the systems side it means ROS 2, real-time control, sensor fusion, and the unglamorous work of making a model fast enough to run on an embedded GPU.
The interesting problems live where those two halves meet: distribution shift between simulation and reality, latency budgets that make a great model unusable, and calibrating how much a planner should trust a network that is confidently wrong. I care about measurable results and honest failure analysis.
- X+
- Years in robotics & AI
- XX
- Models trained & deployed
- XX
- Systems shipped to the field
المشاريع
Language-Conditioned Robot Manipulation
2025Natural-language instructions grounded into executable robot plans, without per-object training.
- PyTorch
- LLMs
- CLIP
- Open-vocab detection
- ROS 2
- MoveIt 2
- TensorRT
حزمة ملاحة الروبوت المتنقل
2025A full ROS 2 autonomy stack for a warehouse AMR, with a learned traversability costmap replacing hand-tuned heuristics.
- ROS 2
- C++
- Nav2
- SLAM Toolbox
- PyTorch
- PREEMPT_RT
- Docker
Real-Time Detection & Tracking on the Edge
2024Multi-object detection and tracking fast enough to sit inside the control loop on a Jetson.
- YOLO
- ByteTrack
- TensorRT
- CUDA
- Jetson
- ONNX
Sim-to-Real Deep Reinforcement Learning
2024A locomotion policy trained in simulation that survived terrain the hand-tuned baseline could not.
- PPO
- Isaac Sim
- PyTorch
- MuJoCo
- Domain randomization
Visual-Inertial SLAM
2024Tightly-coupled visual-inertial odometry with a factor-graph back end for GPS-denied flight.
- C++
- OpenCV
- GTSAM
- Ceres
- Eigen
Retrieval-Augmented Technical Assistant
2023A RAG system answering questions over technical documentation, evaluated against human labels.
- Transformers
- LoRA
- RAG
- Vector search
- FastAPI
- Docker
Skills
Languages
- Python
- C++17 / 20
- CUDA
- SQL
- Bash
- CMake
Machine Learning
- scikit-learn
- XGBoost / LightGBM
- Feature engineering
- Hyperparameter search (Optuna)
- Model evaluation & calibration
- Time-series forecasting
Deep Learning
- PyTorch
- TensorFlow / Keras
- CNNs & Vision Transformers
- Reinforcement learning (PPO / SAC)
- Distributed & mixed-precision training
- Model compression
Computer Vision
- OpenCV
- Detection (YOLO / DETR)
- Semantic & instance segmentation
- Stereo & depth estimation
- Visual SLAM / VIO
- Camera calibration
NLP & LLMs
- Hugging Face Transformers
- Fine-tuning (LoRA / QLoRA)
- RAG & vector search
- Embeddings & semantic search
- Tokenization, spaCy / NLTK
- Evaluation & guardrails
Robotics
- ROS 2 (Humble / Jazzy)
- Nav2 & MoveIt 2
- Sensor fusion (EKF / UKF)
- Motion planning
- Real-time control (MPC / PID)
- Behavior trees
Simulation
- NVIDIA Isaac Sim
- Gazebo / Ignition
- MuJoCo
- Domain randomization
- Sim-to-real transfer
- Foxglove / rviz2
MLOps & Deployment
- Docker & CI/CD
- Weights & Biases, MLflow
- ONNX / TensorRT
- Jetson & edge inference
- Data & experiment versioning
- Monitoring & drift detection
Embedded & Hardware
- Linux PREEMPT_RT
- STM32 / FreeRTOS
- CAN / CANopen
- LiDAR, IMU, depth cameras
- Time synchronization
- Hardware-in-the-loop
Experience
Senior Robotics & AI Engineer · Company Name
2023 — PresentPerception and autonomy for a fleet of N robots deployed across M customer sites.
- Replaced a hand-tuned perception heuristic with a learned segmentation model, lifting obstacle recall from XX% to YY% while holding the inference budget under Z ms on embedded hardware.
- Owned the training-to-deployment path: dataset curation, active-learning loop on field failures, TensorRT export, and shadow-mode evaluation before promotion.
- Built a hardware-in-the-loop CI rig that replays field recordings on every merge — caught N regressions before they reached a robot.
- Mentored N engineers across the ML and robotics sides of the team.
Computer Vision & Machine Learning Engineer · Previous Company
2021 — 2023Vision models for automated inspection and robotic manipulation.
- Trained and shipped detection and segmentation models that raised defect catch rate from XX% to YY% at a Z% false-positive rate.
- Cut inference latency X× through quantization and operator fusion, moving the model from a server GPU to an on-device accelerator.
- Built the labelling and evaluation pipeline — including the class-imbalance handling that made rare-defect metrics trustworthy.
Robotics Engineer · First Company
2020 — 2021Early-stage team building an outdoor inspection platform.
- Developed sensor drivers and time-synchronization for a LiDAR + IMU + stereo rig.
- Prototyped the teleoperation stack and the first autonomous waypoint-following demo.
Education
- B.Sc. Computer / Electrical Engineering · University Name2014 — 2018
Publications
Contact
Let\u2019s work together.
I\u2019m open to full-time roles, contract work, and interesting collaborations \u2014 especially anything involving autonomy in the real world. Email is the fastest way to reach me.
- Based in
- Aleppo, Syria
- Response time
- Within a couple of days