Available remote · Cairo, UTC+2 · 20–25 h/wk now, full time from mid-2027

Robotics engineer.Perception, localization and control on real hardware.

ROS 2, computer vision and sensor fusion, deployed on the hardware that actually flies and drives: Jetson, STM32, Raspberry Pi. Bachelor's thesis graded 1.0 (A) at the University of Stuttgart's Institute of Flight Mechanics and Controls. Delivering GNSS/RTK localization, photogrammetry pipelines and vision systems as a remote contractor.

Mostafa Abdelaziz holding his bachelor thesis in front of the Universität Stuttgart building
Thesis hand-in · Universität Stuttgart, 2026Institute of Flight Mechanics and Controls, graded 1.0 (A)
SCROLL
Track record
0
projects shipped across robotics, perception and embedded systems, from flight controllers to photogrammetry pipelines.
Egypt · Germany
0
companies and labs worked at: Valeo, Lucas-Nülle and the Multi-Robot Systems Lab, in Cairo and Germany.
Client reviews · Upwork
0 × 5.0
five-star client reviews, every contract rated 5.0: ROS 2, EKF fusion, embedded firmware, 3D tooling. Read them ↓
University of Stuttgart
0
thesis grade at the Institute of Flight Mechanics and Controls, the highest grade on the German scale.
German University in Cairo
0
rank in Mechatronics by cumulative GPA, 0.93 (≈ 4.0). GUC scholarship awarded on that ranking; DAAD scholarship for a semester in Berlin.
Hands on hardware
0
robot platforms: blimp, quadcopter, 4-drone swarm, Unitree Go1, TurtleBot 3, Franka Panda, RTK rover, Jetson robot arm.
Selected work

Eleven projects, measured not described.

Numbers and figures come from each project's own repository, logs or outputs.

The real research airship in flight during an outdoor experimentthe real airship in flight · outdoor experiment, FRPG / Perceiving Systems
deployment model in the interactive 3D viewer
prediction vs ground truth
The airship flying over the test field, watched from the groundover the test field · field trial
recursive rollout, straight-line phase
Thesis · graded 1.0 (A)PyTorchPhysics-informed

PR-TCN: physics-residual motion prediction for airships

Institute of Flight Mechanics and Controls · University of Stuttgart · 2026

A docking controller needs to know how a blimp will react to a gust before it happens. The prior predictor handled one vehicle in straight flight. I extended it to 17 manoeuvres, five vehicle sizes and four wind models, and replaced the purely learned model with a frozen analytical rollout plus a gust-conditioned TCN that learns only the residual.

The finding I care about most is a negative one: the most accurate analytical prior is not the best one to learn on top of. Of ten architectures across eight held-out shifts, the physics-residual model is the only one that beats the baseline on every shift.

−88.6%
velocity error vs constant-velocity fallback
4.8 cm
position error at 1.5 s horizon, vs ~24 cm
519.8 h
flight recorded · 31,300 episodes released
54
real-flight episodes · up to −60% position error
10 Hz
control budget met on embedded hardware
Temporal conv netsGazeboROSBayesian HPOEmbedded inference

Supervised by Pascal Goldschmid, examined by Jun.-Prof. Dr.-Ing. Aamir Ahmad. Builds on arXiv:2511.19135. Airship photos: Flight Robotics and Perception Group (iFR, University of Stuttgart) / Perceiving Systems (MPI-IS), from Perception-driven Formation Control of Airships (Price, Black, Ahmad) and the Airship-MPC repository. The 1.0 is the top grade on the German scale (A).

RTK localization on the rover: GNSS satellites, an RTK base station on a tripod sending corrections, and the dual-antenna rover driving a waypoint route
loading models RTK localization on the rover · drag to orbit
Client contractROS 2RTK GNSSField robotics

RTK GNSS localization & waypoint navigation for a field rover

ROS 2 Humble on a Jetson · dual-antenna RTK receiver · robot_localization EKF · shipped and field-validated, remotely

Integrated and hardened the localization chain of an autonomous outdoor rover: a dual-antenna RTK GNSS receiver feeding navsat_transform and a 50 Hz robot_localization EKF that fuses wheel odometry with centimetre-level RTK position, driving go-to-waypoint and go-straight controllers over an RS485 motor driver, with twist_mux arbitration between autonomy and teleop.

Because valid field time was the scarce resource, I built a hardware-in-the-loop simulator that runs the rover's real ROS 2 stack (only the receiver and motor driver are replaced by models) and validated it against the field record on three independent axes before trusting it. Field campaigns were analysed from the logs; every integration defect found (a drivetrain channel convention, a heading declination bias in navsat_transform, an arming gate starved by a 1 Hz heading topic) shipped with an offline replay that proves the change against the recorded RTK ground truth, unit tests against a plant model, a reviewable patch and a scripted field checkpoint. I also wrote the client team's three-tier GNSS/RTK reference.

RTK positioning quality measured from the field logs: centimetre-level scatter of fixes at rest, and a logged run's RTK ground track at 10 Hzfrom the field logs · fixes at rest, a 10 Hz run
50 Hz
EKF fusing wheel odometry + RTK position
3 mm
95 % RTK repeatability at rest, 10 Hz fixes
3 / 3
replay windows agree with RTK after integration
0
site visits; everything delivered remotely
robot_localizationnavsat_transformtwist_muxHIL simulatorField-log replayPythonGNSS reference authored

Client project under NDA. Plot computed from the field logs, relative coordinates only. Illustration rendered from public 3D models: NASA (satellite), Clearpath Robotics (Husky), HELIOS++ (tripod).

EKF velocity, heading and yaw rate against GPS-derived speed, GPS course and raw gyro for a 42-second driving sessionraw sensors vs fused estimate · session 77, 458 m at ~40 km/h
Route map: raw 1 Hz GPS fixes and the fused EKF positionroute · raw 1 Hz GPS fixes vs fused position at IMU rate
Client contractC++15-state EKF

15-state EKF for smartphone GNSS + IMU fusion

C++ core with Python and C bindings · batch and online modes · cars and motorcycles

A sensor-fusion library that turns a phone's raw GNSS fixes and IMU stream into vehicle kinematics at IMU rate: velocity, tangential acceleration, heading, pitch, roll (lean angle for motorcycles), yaw rate and corrected position. What makes it work in practice is the automatic calibration of the phone's arbitrary mounting orientation into the vehicle frame, estimated from an ordinary driving session through gravity and velocity alignment rather than a rig. A later engagement on the same system removed the slope effect from the acceleration estimate, which the client describes in the reviews below. Runs on recorded files or on a live stream.

On a recorded driving session, the fused velocity tracks speed derived from consecutive GPS fixes with r = 0.95, and heading converges within five seconds to a median error of 0.4° (RMSE 1.3°) against GPS course over ground, using the fixes as reference because the phone's own speed field is a placeholder with corrupt spikes.

r = 0.95
EKF velocity vs GPS-derived speed · RMSE 1.19 m/s
0.4°
median heading error after convergence · RMSE 1.3°
234 Hz
fused output from 1 Hz GPS + IMU
GNSSIMUMount auto-calibrationEigenPython & C bindings

Figures from the repository's own input and output files.

four drones following a figure-8 trajectory in formation · simulation
Client contractROS 2PX4Gazebo

Multi-drone formation flight

ROS 2 · PX4 SITL · Gazebo · C++ control · MPC planning

A four-drone formation-flight system built on ROS 2 with PX4 software-in-the-loop, flying synchronised trajectories such as the figure-8 shown, with model-predictive planning, a C++ controller, collision avoidance between the drones and an enforced 3D geofence, tested under wind disturbance.

4
drones in configurable formations
MPC
receding-horizon planning
5 m/s
wind disturbance tolerated
Multi-robotPX4GeofencingC++

Client project under NDA.

One of the 85 input phone photos: the cookie on a wooden stump, outdoors, background visible 1input photo · 1 of 85 phone camera, handheld · stage-1 mask overlay dims the background
2point cloud
loading… drag to orbit · scroll to zoom
3textured GLB · 0.76 MB 120,000 triangles · Draco + WebP drag to orbit · scroll to zoom
PhotogrammetrySAM 2.1COLMAPOpenMVSInteractive

dish-to-glb: phone photos to a web-ready 3D model

Nine stages, one command, every number measured · the model above is the actual pipeline output

85 handheld phone photos in, a 0.76 MB Draco+WebP GLB out: object only, no table, no floaters, one connected component. 31.9 minutes on a CPU-only laptop. Metric scale from a printed, calipered ChArUco board.

The finding: background isolation is the main per-scan cost control, but where you apply it decides whether the scan works. Masking during alignment starved registration (60/85 vs 83/85 images); masking during densification confined depth estimation to ~13% of pixels. Align unmasked, densify masked.

83 / 85
images registered · 0.82 px reprojection
120k
triangles in the final mesh
31.9 min
end to end, CPU only, no CUDA
0.76 MB
final GLB · 7.05 MB USDZ
GLB / USDZDracoChArUcoBlenderthree.jsPython

Panel 2 samples 200,000 points on the reconstructed surface in your browser, coloured from the texture or by elevation; panel 3 is the pipeline's GLB. Input photos: Epic's public RealityScan sample set.

Mostafa with the Unitree Go1 quadruped in the IAS lab at the University of Stuttgartwith the Go1 · IAS lab, University of Stuttgart · Nov 2024
Legged robotROSGazeboPath planning

Autonomous navigation & path planning for the Unitree Go1

Robotics & ROS engineer · fully funded workshop, IAS, University of Stuttgart

Developed navigation and path-planning for the Unitree Go1 quadruped as part of a fully funded workshop at the University of Stuttgart's IAS: waypoint tracking, obstacle avoidance and autonomous mobility in ROS and Gazebo, using an artificial-potential-field planner. Improved the robot's ability to navigate dynamic environments and contributed the results to a DAAD-supported multi-robot cooperation project alongside TurtleBot 3 and Franka Emika Panda platforms.

Go1
Unitree quadruped, real hardware
3
heterogeneous robot platforms integrated
DAAD
funded exchange, presented to industry
Artificial potential fieldsObstacle avoidanceWaypoint trackingMulti-robotPython
RViz view of the SLAM map being built in simulation: laser scan, costmap and the robot's TF framesRViz · slam_toolbox map + laser scan, in simulation
TF frame tree of the robot: map, odom, base_link, wheels, chassis, camera and lidar frames with publish ratesTF tree recorded from the running system (view_frames)
Client contractROS 2 JazzyNav2slam_toolboxSimulation

UV-C disinfection robot: autonomy stack in simulation

Gazebo Harmonic · differential drive · human-safety interlocks · 262 passing tests

The autonomy software for an indoor UV-C disinfection robot, built and validated in simulation: the robot maps a room with slam_toolbox, plans an optimised route between disinfection waypoints with Nav2, holds the lamps on for configurable dwell times, and shuts them down immediately on human detection, because the payload is genuinely dangerous. My work was the simulation and software stack (mapping, navigation, route optimisation, lamp control and safety), with the hardware deployment path documented.

262
passing tests across the stack
37.7 Hz
odom → base_link, from the recorded TF tree
DWB
Nav2 local planner
Gazebo HarmonicRoute optimisationSafety systemsPython
Source on GitHub
laser following
circle trajectory
catch & stabilise
The 2-DOF platformthe rig
JavaFX tuning and telemetry GUIJavaFX live-tuning GUI
STM32OpenCVReal-time control

Vision-guided 2-DOF ball balancing platform

Full stack: optics → firmware → controller → GUI

A ball is balanced and driven along arbitrary trajectories on a tilting plate by an STM32F103C8T6 closing the loop on positions from an OpenCV pipeline. Three controllers are switchable at runtime (PID, PV and LQR), with a Kalman filter for velocity and an EMA on the control output. Gains are tuned live over serial while the plate is running.

3
controllers, comparable live
2-DOF
MG996R servo tilt platform
Live
serial tuning & telemetry
Embedded CFreeRTOSKalman filterPID / LQRJavaFXUART · TCP/IP
Source & full report
The DroneX52 quadcopter next to its FlySky transmitterthe airframe · STM32 flight controller, FlySky transmitter
The quadcopter hovering during a night test flighthover test · night flight
Ground station: live PID tuning and real-time pitch, roll, yaw and hover plotsground station · live PID tuning, desired vs actual pitch / roll / yaw / hover
STM32Flight controlSensor fusion

DroneX52: an STM32 quadcopter with a custom flight controller

Bare-metal STM32F103 · PID stabilisation · IMU + barometer fusion · Multi-Robot Systems Research Group, GUC · 2024

A quadcopter built from the frame up around an STM32F103 flight controller: PID loops on pitch, roll and yaw driven by fused gyroscope, accelerometer and barometer data, UART links for Bluetooth and WiFi control, and a hand-motion mode flown from the smart glove further down this page. It flies: the night hover above is a real test flight.

The ground station is the part that made tuning tractable: a Java GUI plots desired against actual pitch, roll, yaw and hover in real time and pushes PID gains to the controller live, so gains were tuned in the air rather than by recompiling firmware.

3
PID loops: pitch, roll, yaw
3
sensors fused: gyro, accel, baro
Live
gain tuning over UART
STM32F103Embedded CPIDIMUJava GUIUART · Bluetooth · WiFi
Source on GitHub
The ESP32-S3 touch alarm clock running the firmwarethe device, running the firmware
The clock's touch display showing the alarm interfaceLVGL touch UI
The clock hardwareESP32-S3 3.5" touch board
Client contractFirmwareESP32-S3ESP-IDFLVGL

Zmanim alarm clock: full firmware for an ESP32-S3 touch display

Off-the-shelf ESP32-S3 3.5" touch board · every line of software mine: drivers, UI, alarm engine, solar-time engine, simulator

Firmware for an alarm clock that understands zmanim, the halachic times derived from sunrise and sunset. A NOAA solar calculator and a bundled world-city database (coordinates, elevation, time zone) compute the day's times on-device, so an alarm can be set as "10 minutes before Shma" and it moves with the calendar; alarms carry day-of-week filters and offsets, with a test suite covering offset maths, 12/24-hour conversion, time-zone handling and edge cases.

Around that engine: up to eight labelled alarms and a countdown timer, an LVGL 8 touch interface at 60 FPS, MP3/WAV alarm sounds through the board's I2S DAC, a battery-backed RTC, settings persisted in NVS and a sub-5 mA sleep mode, with the ST7789 display, CST328 touch, PCM5101 audio and PCF85063 RTC drivers written against ESP-IDF 5. A Windows LVGL simulator runs the same UI on a PC for development, and a hardware-abstraction layer let the shared UI and alarm code be ported to the ESP32 "CYD" board.

8
alarms + timer, persisted in NVS
60 FPS
LVGL UI · <50 ms touch response
<5 mA
standby · ~1.4 MB flash footprint
ESP-IDF 5LVGL 8FreeRTOSI2S audio · libhelix-mp3PCF85063 RTCNOAA solar calcPC simulatorC
Training and validation curvestraining & validation curves
Precision-recall curveprecision–recall
Normalised confusion matrixconfusion matrix
PerceptionYOLOv8Custom training

Custom-trained YOLOv8 ball & laser detector

Classical CV benchmarked against a trained model, switchable at runtime

The HSV pipeline was fast but fragile under changing light. I built the full loop (capture, annotation, training, evaluation) for a custom YOLOv8-nano detector, and kept both methods live so the trade-off is measurable rather than asserted.

UltralyticsOpenCVDataset annotationPython
Source on GitHub
Client reviews

Twelve contracts, twelve five-star reviews.

Screenshots from my Upwork profile.

Upwork

Robotics sub-contract

Upwork · Jan 20, 2026
5.0

"His code always worked out of the box and I didn't have to make several iterations."

Five-star Upwork review: more of an employee than a sub-contractor, cared that the whole system works, responsive even at weekends, code worked out of the box, highly recommended
Upwork

Debug driver actions (EKF fix)

Upwork · Jun 15 – Jul 18, 2026
5.0

"He knew both to implement the theory and where patches should be added in order to make it work."

Five-star Upwork review: fixed a problem in our EKF by cancelling the slope effect on acceleration estimation; fast, high-quality solutions with deep understanding of theory and project need
Upwork

ROS 2 programming for a differential-drive mobile robot

Upwork · Jan 8 – 15, 2026
5.0

"Mostafa met a very tight deadline and managed the work perfectly."

Five-star Upwork review: met a very tight deadline, strong technical expertise, excellent problem solving, highly reliable and proactive
Upwork

ESP32 touchscreen alarm clock with custom features

Upwork · Nov 17, 2025 – Jan 23, 2026
5.0

"Mostafa really treated this project like his own. He went above and beyond expectations."

Five-star Upwork review: treated the project like his own, went above and beyond, added many features, highly recommended
Upwork

C++ audio capture and encoding app

Upwork · Jan 2 – 8, 2026
5.0

"Knows what he's doing unlike 80% of people here that are willing to accept a job they have no idea about."

Five-star Upwork review: amazingly professional job, well-documented code, delivered very fast
Upwork

ESP-Home + StackChan firmware integration

Upwork · Aug 5 – 17, 2026
5.0

"The best thing about working with Mustafa is that he is very clear and honest."

Five-star Upwork review: reliable, clear and honest, completed on time, always available for extra support
Upwork

STL desktop tool: mirror orthoses, engrave logo and patient text

Upwork · Jan 22 – Feb 2, 2026 · repeat client
5.0

"This is already the second time I have worked with Mostafa. As usual, everything worked out very well."

Five-star Upwork review from a repeat client: clear communication, reliable work, delivered on time, fully recommended
Upwork

LadybugCapPro SDK command-line tool

Upwork · Feb 3 – 11, 2026
5.0

"He delivered promptly, met all requirements, and provided a correct and reliable solution."

Five-star Upwork review: delivered promptly, met all requirements, strong technical skills and professional communication
Upwork

Contract review

Upwork · May 15, 2026
5.0

"Rare talent - intelligent, hardworking and committed."

Five-star Upwork review: rare talent, intelligent, hardworking and committed
Upwork

Contract review

Upwork
5.0

"Very communicative, professional and committed for the project success beyond his part."

Five-star Upwork review: communicative, professional and committed to project success beyond his part
Upwork

Contract review

Upwork
5.0

"A very intelligent, hardworking and committed individual."

Five-star Upwork review: very intelligent, hardworking and committed
Upwork

3D desktop application for custom insole adaptation

Upwork · Jan 8 – 14, 2026
5.0

Endorsed: Solution Oriented · Clear Communicator · Accountable for Outcomes

Five-star Upwork review with endorsements: solution oriented, clear communicator, accountable for outcomes
Also built

Range across the whole stack.

Scale-model intersection with working traffic lightsclick to enlarge

Smart traffic system on FPGA

Full traffic-light controller in VHDL on a Basys3, driving a physical intersection model: state machines, timing and hardware debounce. See the wiring ↗

The glove with flex sensors wired to an ATmega32 development boardclick to enlarge

Smart glove for the deaf

ATmega32 wearable that reads a flex sensor on each finger and translates hand gestures into text on an LCD and over Bluetooth.

Dice Realms game board screenshotclick to enlarge

Dice Realms

Led six engineers on a JavaFX dice-drafting strategy game with hot-seat multiplayer and an AI opponent that placed 1st among ten teams in a university tournament. Wizard select ↗

Stack-chan robot: an M5Stack display module with a face on a two-servo body, RT Corporation versionclick to enlarge

StackChan × Home Assistant

One ESP32-S3 firmware that exposes the M5Stack StackChan robot (head servos, RGB bars, display, camera, speaker) as auto-discovered Home Assistant entities over MQTT, while the on-device AI agent keeps working and can control the home by voice. Verified on hardware. See the architecture ↗

Orthosis Customizer desktop app: engraving controls on the left, left and right orthosis 3D viewsclick to enlarge

Orthosis Customizer

Windows desktop app that engraves a logo, patient name and date into an orthosis STL, then exports matched left and right versions: trimesh-based processing, pick-on-mesh engraving position, and a side-by-side 3D viewer, shipped as an installer.

Source excerpt from the AES assembly implementationsource excerpt

AES block cipher in Java & 8086 assembly

Led a team of three implementing the Advanced Encryption Standard: a complete Java implementation (SubBytes, ShiftRows, MixColumns, round keys) and an 8086 assembly port in emu8086 carrying the S-box, round-constant and Galois-field tables and the round structure.

Attendance management system, Courses tab: course creation, batch setup, a sessions table with dates and times, and an export to .xlsx buttonclick to enlarge

Attendance management system

Scalable attendance system for training centres and academic institutions: a JavaFX front end over a MySQL database: courses, batches, sessions and per-session attendance.

Skills & tools

What I build with, by layer.

By layer, from simulation down to the silicon.

Robotics & simulation

ROS 2 Humble · Jazzy Nav2 MoveIt 2 RViz slam_toolbox Gazebo Webots AIRAirSim ArduPilot SITL Mission Planner PX4 AeroStack2 Isaac Sim

Perception & 3D

OpenCV Ultralytics YOLO Segment Anything 2 COLMAP MVSOpenMVS Blender three.js

Machine learning & data

PyTorch CUDA NumPy pandas SciPy Jupyter

Embedded & edge

NVIDIA Jetson Orin Nano Raspberry Pi STM32 ESP32 ESP-IDF Arm Cortex-M FreeRTOS LVGL AVR ATmega32 Linux Home Assistant

Languages

Python C++ C Java JavaFX MATLAB VHDL Xilinx Bash LaTeX

Tooling & design

Git GitHub CMake Ubuntu MySQL SolidWorks
Background

Mechatronics, with the practice done in Germany.

Mostafa receiving an award certificate at the German University in Cairo

Awards & recognition

GUC scholarship award ceremony · German University in Cairo
  • GUC scholarship, sponsored by the German University in Cairo on the strength of cumulative ranking in Mechatronics Engineering.
  • DAAD Scholarship Award: the competitive DAAD scholarship, administered by Ulm University, for outstanding academic performance as a top-ranked Mechatronics student at GUC.
  • Ranked 3rd in Mechatronics by cumulative GPA at the German University in Cairo: 0.93, equivalent to 4.0.
  • Bachelor thesis graded 1.0 (A) at the University of Stuttgart's Institute of Flight Mechanics and Controls.
  • 1st place: the Dice Realms AI player won a university-level tournament against ten teams.
  • 1st in school: Futures Language School, 97.1% in the mathematics section.
05/2026 – 08/2026

Robotics & AI Werkstudent

Lucas-Nülle GmbH · Cologne, Germany

Commissioned a ROS 2 Humble robot-arm platform on a Jetson Orin Nano: a depth camera scanning the workspace into a live 3D model in RViz, a bidirectional MoveIt 2 digital twin between simulation and hardware, GPU-accelerated object detection with autonomous sorting, and a PyQt5 teaching app with a Blockly interface. Built an ArduPilot-based software-in-the-loop drone training environment: quadcopter, fixed-wing, flying-wing and VTOL models, a georeferenced terrain pipeline with a local WMS, and a library of 20 reproducible fault-injection scenarios, packaged for offline deployment. Reference letter ↓

03/2026 – 07/2026

Bachelor thesis, graded 1.0 (A)

Institute of Flight Mechanics and Controls (iFR) · University of Stuttgart

"Extending a TCN-Based Blimp Motion Predictor to Versatile Vehicle and Flight Conditions." Faculty of Aerospace Engineering and Geodesy.

07/2025 – 09/2025

Embedded Systems Intern

Valeo · Cairo, Egypt

System testing for embedded automotive projects. Built an automated test-management tool that raised throughput by over 200%.

11/2024

DAAD-funded exchange workshop

IAS · University of Stuttgart

Represented GUC at "Cooperation of Heterogeneous Agents in Industrial Applications", integrating TurtleBot 3, Unitree Go1 and Franka Emika Panda. Specialised in ROS, Gazebo and Go1 navigation.

08/2024

Mechatronics Research Intern

Multi-Robot Systems (MRS) Research Group · German University in Cairo

Developed an STM32-based quadcopter with PID control.

12/2023 – 06/2024

Semester abroad on a DAAD scholarship

German International University · Berlin

Awarded a DAAD scholarship for academic excellence for the fourth semester in Berlin.

2022 – 2027

BSc Mechatronics Engineering

German University in Cairo · ranked 3rd in Mechatronics by cumulative GPA

Cumulative GPA 0.93 (≈ 4.0). Faculty of Engineering and Materials Science. GUC scholarship awarded on cumulative ranking.

AMIT Learning
Licence & certification

Embedded Software Engineer: Embedded Systems Diploma

AMIT Learning · issued Mar 2024 · 175 hours · grade 97%

Embedded C and data structures, interfacing protocols, real-time operating systems, testing and validation of embedded systems, automotive bus technology.

AMIT Learning Embedded Systems Diploma certificatecertificate
Lucas-Nülle
Reference letter

Letter of Recommendation from Lucas-Nülle GmbH

Ralf Linnertz, Team Lead Product Management Energy & Industry · Kerpen, 31 Aug 2026 · full letter on request

"Consistently demonstrated a very high level of technical expertise, strong analytical thinking, and exceptional problem-solving skills … performed his duties to our full satisfaction. His conduct toward supervisors, colleagues, and project partners was exemplary at all times."

confidential · available on request
IEEE
Volunteering

Embedded Systems Tutor

IEEE GUC Student Branch · Sep 2024 – Jun 2025 · 10 months

Ran workshops and training sessions teaching engineering students the fundamentals and advanced topics of embedded systems.

Contact

Have a perception, localization or edge problem that's stuck?

I'm in Cairo (UTC+2), a full working-day overlap with Europe and the Gulf, mornings with the US east coast. 20–25 hours a week now through July 2027, full time remote after. Contract, subcontract or employed all work. Tell me what's stuck and I'll tell you honestly whether I'm the right person.