IJAZ AHMED
Interactive 3D LiDAR scan of a street: a ground robot scans buildings, cars, people and trees. Switch the lens to see it as perception, telemetry or segmentation.

Ijaz Ahmed

Robotics · Autonomous Systems · ROS2

Robotics & Systems Engineer

I build autonomous systems that work in the real world — behavior tree mission planners, Nav2 integration, and the backend APIs bridging robots and operators. Based in Mannheim, Germany.

  • ROS2
  • BehaviorTree.CPP
  • Nav2
  • MoveIt
  • Gazebo
  • C++
  • Python
  • FastAPI
  • Docker

Backend · APIs · Systems Integration

Backend Engineer

I design and ship production-grade backend systems — REST APIs, microservices, CI/CD pipelines, and the integration layer between software and physical hardware. 27 delivered projects, 100% success rate.

  • FastAPI
  • Docker
  • Python
  • REST APIs
  • CI/CD
  • ReactJS
  • Node.js
  • GitHub Actions
  • Linux

AI · ML · Computational Science

ML Engineer

I apply machine learning to real engineering problems — materials microstructure modelling, FEM simulation, and computer vision pipelines on autonomous robots. Research-backed, production-minded.

  • PyTorch
  • OpenCV
  • FEM
  • TensorFlow
  • YOLOv8
  • Signal Processing
  • MATLAB
  • Python
See the armory

Working Student @ SAP SE · Walldorf · Mannheim / Remote

Upwork projects · 100% success
27
Years shipping production code
6+
Ornithopter flapping-wing MAV
22g
BT tick rate on deployed UGV
10Hz

// 01 · Hall of Armor

Every mark, running

24 projects across robots, backends and hardware. Every one has a demo running; click any mark to inspect it up close.

SIM

MK 01 · Robotics · ROS2

Autonomous patrol robot with reactive BT mission planning, Nav2, camera-based anomaly detection (OpenCV + YOLOv8), GPS/IMU telemetry, and sim/real switching.
  • ROS2
  • BehaviorTree.CPP
  • Nav2
  • OpenCV
  • YOLOv8
  • FastAPI
  • Docker
SIM

MK 02 · LLM · Evaluation — Open Source

Twelve open-weight models, one identical prompt, 500 opus_books EN–DE pairs at temperature 0, scored with sacreBLEU + chrF. Resumable (append-only JSONL), disk-bounded (pull → translate → delete, so twelve models run on a machine that holds one) and backend-swappable. All 6,000 raw outputs are committed next to their references. Best zero-shot: 13.81 BLEU / 40.09 chrF (qwen3.5:4b).
  • Python
  • Ollama
  • sacreBLEU
  • chrF
  • JSONL
SIM

MK 03 · LLM · Agents — Open Source (MIT)

An LLM agent that notices a capability gap, searches the npm registry, installs an MCP server mid-conversation and uses the new tools in the same turn. Every failed tool call lands in a mistakes table, and the resulting lessons are injected into the system prompt so a failure is not repeated. Parallel tool execution, cached tool schemas. 8,779 LOC TypeScript.
  • TypeScript
  • MCP
  • Node.js
  • LLM agents
SIMLIVE DEMO

MK 04 · Tool · Open Source

Lightweight browser-based visualizer for BehaviorTree.CPP XML files. Live XML editor, pan/zoom canvas, vertical/horizontal layout, minimap, node search. Zero dependencies. Single HTML file. Alternative to Groot2 without any install.
  • HTML Canvas
  • Vanilla JS
  • BT.CPP format 4
  • Zero deps
SIM

MK 05 · SDK · ROS2 · Open Source

Async Python SDK for industrial robot arm control via ROS2. Built from real CR5 and UR5 fiber composite layup work at NUST (2023–2024). Currently migrating from ROS1 Noetic to ROS2 Jazzy — this is v0.1.0 of that migration.
  • ROS2 Jazzy
  • ROS1 Noetic
  • Python
  • Async
  • CR5
  • UR5
SIM

MK 06 · Robotics · ROS2 Jazzy · Learning Path

TurtleBot3 navigation and autonomy stack implemented for ROS2 Jazzy. Part of a structured robotics learning roadmap: TurtleSim → TurtleBot3 → Husky + custom packages → multi-robot cooperation.
  • ROS2 Jazzy
  • TurtleBot3
  • Nav2
  • Python
SIM

MK 07 · Robotics · From Scratch · Open Source

End-to-end robotics project implementations from idea to product. A living collection covering system design, embedded control, ROS2 integration, and deployment — the full journey from concept to working hardware.
  • ROS2
  • Python
  • C++
  • Embedded
SIM

MK 08 · Language Design · Industrial Robotics

A domain-specific language for industrial robot arm control. Readable by non-roboticists. Maps cleanly onto real motion primitives (MOVJ, MOVL). Intent is separate from execution — every motion is kinematically proven before it reaches hardware. Derived from hands-on CR5 and UR5 layup work at NUST.

Intent → Admission → Kinematic Proof → Execution → Supervision
  • DSL
  • Kinematic proof
  • CR5 · UR5
  • Python
SIM

MK 09 · Robotics · ROS1 · Research

Pick-and-place layup sequences with Dobot CR5 and SJ602-A arms in ROS Noetic and Melodic. Multi-robot TCP/IP coordination with MoveIt and MoveGroup. Fixed Noetic compatibility bugs in the official CR5 repository — original code only functional on Melodic. Includes documentation.
  • ROS Noetic
  • ROS Melodic
  • MoveIt
  • MoveGroup
  • Gazebo
  • RViz
  • TCP/IP
SIM

MK 10 · Robotics · ROS1 · Industrial · Legacy

Original ROS1 Noetic codebase for the CR5 fiber composite layup sequences at NUST. Cartesian-space motion planning, MoveIt trajectories, multi-robot TCP/IP coordination. Direct ancestor of the robolink SDK.
  • ROS1 Noetic
  • MoveIt
  • CR5
  • Python
  • TCP/IP
SIM

MK 11 · Robotics · ROS · Simulation

ROS Melodic simulation — TIAGo picks coloured objects from one table, places them on different tables by colour. Variations include: object detection + avoidance with Octomap, multi-room navigation, and Aruco marker detection. A lengthy, complex project covering perception, manipulation, and planning end-to-end.
  • ROS Melodic
  • TIAGo
  • Octomap
  • Aruco
  • MoveIt
  • Object detection
SIM

MK 12 · Competition · Team NUST-AG

Autonomous vehicle competition using the Carla simulator with a ROS bridge. Full navigation pipeline: camera and lidar integration, path planning, vehicle control logic. Carla experience is uncommon and directly relevant to ADAS and simulation engineering roles.
  • ROS
  • Carla
  • LiDAR
  • Computer Vision
  • Path planning
SIM

MK 13 · Competition · Team NUST-AG (Pakistan) · Indonesia

Driver-assistance unit for Team NUST-AG (Pakistan) at Shell Eco-marathon 2024 on the Mandalika Circuit, Indonesia. A Raspberry Pi 4 combines GPS with on-board sensors (gyroscope, IMU) to locate the car on the circuit and tells the driver to pulse or glide for maximum mileage. The pulse/glide points come from a lap simulation of Mandalika. The driver display shows fuel, acceleration, live estimated mileage and the optimal gear. Pit-to-driver messages and remote telemetry (mileage, health metrics, temperature) run over LoRa to a laptop. Built in several variants.
  • Python
  • PyQt
  • Raspberry Pi 4
  • GPS + IMU
  • LoRa
  • Lap simulation
SIMLIVE DEMO

MK 14 · Tool · Deployed · Open Source

Winnow your AI chat history — view, search, and prune exported conversations from Claude and ChatGPT in one offline HTML file. No server, no install, no account. Drop your exported JSON, cut the dead ends, export clean context. Everything runs locally — your data never leaves your machine.
  • HTML
  • Vanilla JS
  • Zero deps
  • Offline
SIM

MK 15 · Hardware · BSc Final Project

Bio-inspired MAV with a flapping wing mechanism — 22g including embedded control electronics. Full mechanical design, fabrication, and MCU-based servo control. Achieved stable controlled flight. At 22g, structural resonance and power density are the dominant engineering constraints.
  • Embedded C
  • Servo control
  • MAV
  • Bio-inspired
SIM

MK 16 · Embedded · Hardware

Two-axis camera-enabled car with WiFi and Bluetooth control. ESP32-CAM module provides live video feed. Control via WiFi and Bluetooth. Two-axis camera gimbal for remote viewing. Designed as a wireless surveillance platform.
  • ESP32-CAM
  • WiFi
  • Bluetooth
  • 2-axis gimbal
  • C++
SIM

MK 17 · Embedded · Hardware

Arduino UNO-controlled motor car receiving instructions from an ESP8266 over WiFi. Demonstrates serial communication between a WiFi-capable module and a microcontroller, and remote command parsing over a local network.
  • Arduino UNO
  • ESP8266
  • WiFi
  • Motor control
  • C++
SIM

MK 18 · Embedded · IoT

Home automation project using an ESP8266 for WiFi-connected light control. Remote switching and scheduling over a local network. A practical introduction to IoT architecture: device firmware, local network protocol, and simple control interface.
  • ESP8266
  • IoT
  • WiFi
  • C++
SIM

MK 19 · Embedded · Hardware

RC 2-wheel drive robot built on Arduino UNO. Wireless control via radio module. PWM motor speed control and steering logic. One of the foundational embedded projects that led to more complex autonomous systems work.
  • Arduino UNO
  • RF control
  • PWM
  • Motor driver
  • C++
SIM

MK 20 · Embedded · Robotics

Autonomous line-following robot with IR sensor array and PID-based motor control. Tuned for smooth curve following at speed. A classic control systems problem in hardware — the same PID logic that appears in motor controllers and flight controllers.
  • Arduino
  • IR sensors
  • PID control
  • Motor driver
  • C++
SIM

MK 21 · Tool · CLI · Open Source

Fast C directory tree viewer for Linux and Windows with smart grouping of numbered files. Built for developers who work with large codebases and simulation output directories. Cross-platform, dependency-free, single binary.
  • C
  • CLI
  • Linux
  • Windows
  • Cross-platform
SIM

MK 22 · Full-Stack · AI-powered

Full-stack ESG reporting tool — enter GHG emissions data, visualise scope breakdowns, generate AI-powered reduction strategies, and export a branded PDF report. Built with Next.js 15 (App Router), SQLite, Zod validation, Recharts, and GPT-4o for strategy generation.
  • Next.js 15
  • SQLite
  • Recharts
  • GPT-4o
  • Tailwind
  • react-pdf
Language Recording Spreadsheet screenshot
SCREEN

MK 23 · Tool · Language Learning

Browser-based spreadsheet for language learning and vocabulary recording workflows. Upload a word list, record or play back pronunciations per row, and download individual audio clips. Built for hiring freelancers to record how they pronounce words — each row is a word, each action cell a micro-studio.
  • ReactJS
  • Web Audio API
  • File upload
SIM

MK 24 · Scientific Computing · Upwork

PID and LQR controller implementations, double pendulum simulation, and a multi-signal DSP estimator — delivered as Upwork projects with 4.8–5.0★ ratings. Directly applicable to robotics control, embedded signal processing, and sensor fusion.
  • MATLAB
  • PID
  • LQR
  • DSP
  • Signal processing
  • Control theory

// 02 · Mission log

Where I've shipped

The log adapts to the lens you picked at the top.

  1. Jun 2026 – present

    Working Student — UX Accessibility & Embodied AI Challenge

    SAP SE · Walldorf, Germany

    Coaching participating teams through SAP's Embodied AI & Robotics Challenge and writing the Mission Proposals they compete on — task scenarios, success criteria and the evaluation setup that decides what counts as done.

    UX Accessibility (Product Standards): migrating and documenting SAP's accessibility guidelines from the old wiki to the new Product Standards platform — structured content curation. Also coaches teams in SAP's Embodied AI & Robotics Challenge and writes its Mission Proposals.

    • Embodied AI
    • Robotics
    • Evaluation design
    • Accessibility
    • Accessibility
    • Product Standards
    • Content curation
    • Evaluation design
  2. Feb 2025 – Jul 2025

    Werkstudent — Robotics Software Engineer

    Werkstudent — Software Engineer (Robotics Platform)

    EnQineering GmbH · Remote, Germany

    Full software stack for an autonomous fire-fighting UGV in ROS2 Jazzy. Behavior tree mission layer (BehaviorTree.CPP) — task sequencing, fallback logic, reactive replanning. Nav2 path planning, REST API ↔ ROS2 bridge, Gazebo simulation with sim/real mode switching. Code under NDA.

    Backend integration layer for an autonomous ground vehicle — REST API bridge between mission control and the ROS2 runtime, JWT-authenticated mission dispatch, WebSocket telemetry, ReactJS operator dashboard. CI/CD pipelines and automated testing infrastructure.

    Computer vision pipeline on a deployed UGV — camera-based fire detection feeding into a ROS2 behavior tree via OpenCV. Sensor fusion layer combining GPS, IMU, and lidar into the Nav2 costmap. Real-time inference constraints on embedded hardware.

    • ROS2
    • BehaviorTree.CPP
    • Nav2
    • FastAPI
    • Gazebo
    • OpenCV
    • MQTT
    • FastAPI
    • ReactJS
    • Docker
    • JWT
    • CI/CD
    • ROS2 bridge
    • OpenCV
    • ROS2
    • Sensor fusion
    • GPS/IMU
    • Nav2
  3. Nov 2023 – Mar 2024

    Research Assistant — Robotics

    NUST · Islamabad, Pakistan

    Robotic arm manipulation with UR5, Dobot CR5, and SJ602-A. Pick-and-place sequences, multi-robot TCP/IP coordination, trajectory planning with MoveIt. Fixed ROS Noetic compatibility issues in the official CR5 repository (original code only worked on Melodic). Proposed and partially implemented an interface redesign for the SJ602-A. Taught ROS workshops to 20+ students.

    • ROS1/2
    • MoveIt
    • RViz
    • TCP/IP
    • Gazebo
    • Python
  4. Jan 2022 – Jan 2023

    Full Stack Developer

    Klero Généalogie · Remote, France

    Built the backend from scratch for a genealogy research platform — data mining infrastructure with rotating proxy systems for public historical registries at scale, REST + GraphQL APIs, ReactJS frontend, CI/CD pipelines.

    • Python
    • Node.js
    • ReactJS
    • GraphQL
    • REST
    • CI/CD
  5. 2015 – 2021

    Freelance Software Engineer

    Upwork · Remote

    27 completed projects, 100% Job Success Score. Automation, scientific computing, full-stack. PID/LQR controllers in MATLAB (4.8★), multi-signal DSP estimator, gRPC client in Java (5.0★), browser automation suites for UK clients, data scraping pipelines.

    • Python
    • MATLAB
    • Selenium
    • Playwright
    • gRPC
    • 100% JSS

// 03 · Armory

Stack & toolchain

Highlighted items are daily drivers.

Robotics core

  • ROS2 Jazzy
  • Nav2
  • BehaviorTree.CPP
  • ROS Noetic/Melodic
  • MoveIt
  • Gazebo
  • RViz
  • Carla
  • PDDL
  • MQTT

Perception

  • OpenCV
  • YOLOv8
  • PyTorch
  • Octomap
  • Aruco markers
  • LiDAR
  • Sensor fusion

Languages

  • Python
  • C++
  • Bash
  • MATLAB

Backend bridge

  • FastAPI
  • Docker
  • REST APIs
  • WebSocket
  • JWT

Embedded

  • STM32
  • ESP32
  • ESP8266
  • FreeRTOS
  • Arduino
  • CAN bus
  • TCP/IP

DevOps

  • GitHub Actions
  • CI/CD
  • Linux Admin
  • Grafana
  • Jenkins

Backend

  • FastAPI
  • Flask
  • Node.js
  • REST APIs
  • GraphQL
  • JWT
  • WebSocket
  • Next.js

DevOps & Infra

  • Docker
  • GitHub Actions
  • CI/CD
  • Jenkins
  • Linux Admin
  • Grafana

Languages

  • Python
  • JavaScript
  • C++
  • Bash
  • C#
  • Java

Frontend

  • ReactJS
  • Next.js
  • Python Qt
  • HTML/CSS
  • Canvas API

Automation

  • Playwright
  • Selenium
  • Puppeteer
  • gRPC
  • Proxy rotation
  • Multithreading

Integration

  • ROS2 bridge
  • MQTT
  • IIoT
  • TCP/IP
  • SQLite

ML frameworks

  • PyTorch
  • TensorFlow
  • CNN
  • MLP
  • KNN
  • Transfer learning

Computer vision

  • OpenCV
  • YOLOv8
  • Object detection
  • Aruco markers
  • ROS2 image pipeline

Scientific computing

  • MATLAB
  • FEM
  • Microstructure ML
  • PID / LQR
  • DSP / Signal processing

Materials science

  • Microstructure modelling
  • Semiconductor physics
  • Thermodynamics
  • Phase modelling

Deployment

  • FastAPI
  • Docker
  • Python
  • GitHub Actions

Sensors & data

  • Sensor fusion
  • GPS / IMU
  • LiDAR
  • Camera pipelines

// 04 · Pilot

Building things that work

Robotics software engineer and master's student at TU Bergakademie Freiberg, specialising in Computational Materials Science — ML for materials and finite element simulation.

My work sits at the intersection of autonomous systems and backend infrastructure — mission logic that decides what a robot does, and the APIs that let humans monitor it in real time. Most recently at EnQineering GmbH: full UGV software stack in ROS2, behavior trees, Nav2, REST/ROS2 bridge, Gazebo simulation tools.

Backend engineer with 6+ years of shipping production systems — genealogy platform data infrastructure, IIoT dashboards for industrial robots, and a REST/ROS2 integration layer for an autonomous UGV. 27 Upwork projects, 100% Job Success Score.

I'm comfortable with the full backend stack and particularly strong at the hardware-software boundary where most backend engineers stop — writing the API layer that a physical robot actually talks to.

ML engineer and master's student at TU Bergakademie Freiberg — research focus on machine learning for materials science, training models on microstructure data and integrating ML with FEM pipelines.

On the applied side: computer vision for autonomous robotics — OpenCV anomaly detection and YOLOv8 on ROS2, sensor fusion (GPS + IMU), and signal processing (PID/LQR controllers, DSP estimators) from applied Upwork work.

location
Mannheim (Rhein-Neckar), Germany
english
C1
german
B2 (exam this semester)
hours
20h/wk · 40h/wk in breaks
github
iameijaz

Education

  • M.Sc. Computational Materials Science

    TU Bergakademie Freiberg · In progress

    Solid State Physics · Semiconductors · Thermodynamics · Finite Element Methods · ML for Materials · Microstructure Modelling.

  • B.Sc. Mechanical Engineering

    NUST · Islamabad · 2023

    Control systems · Robotics · Mechatronics · Fluid dynamics. Final project: 22g flapping wing ornithopter. Shell Autonomous Programming · Shell Eco Marathon international competitions.

Certifications

  • IBM Data Science Professional Certificate
  • NVIDIA Introduction to Networking
  • MathWorks Systems Engineering
  • Agile with Atlassian Jira

// 05 · Suit up

Let's work together

Currently a Working Student at SAP SE in Walldorf, based in Mannheim. Open to conversations about robotics, embodied-AI and LLM-evaluation work — Rhein-Neckar, Karlsruhe or remote.

Working Student @ SAP SE · open to conversations