# Abdalrahman Ibrahim Head of AI Products Development · AGILOX · Linz, Austria Head of AI Products Development at AGILOX. Builds private LLM infrastructure and on-robot AI for autonomous mobile robots. PhD candidate at the University of Klagenfurt on Agentic AI for Robotics and Transportation. *Open to consulting, advisory, and full-time opportunities.* --- ## §01 Work ### YOLOs-CPP **Creator & Maintainer · 2025 · Active** Header-only C++ library for real-time YOLO inference — detection, segmentation, pose, OBB — no Python, no runtime bloat **Stack:** C++17, ONNX Runtime, OpenCV, CUDA, YOLOv5–v12, Quantization (INT8/FP16) **GitHub:** **Demo:** **Detail:** **Markdown:** --- ### YOLOs-CPP-TensorRT **Creator & Maintainer · 2026 · Active** Header-only C++ YOLO library for NVIDIA TensorRT — GPU preprocessing, CUDA Graph replay, sub-2ms latency, 530+ FPS **Stack:** C++17, TensorRT ≥10, CUDA ≥12, OpenCV, CUDA Graph **GitHub:** **Detail:** **Markdown:** --- ### ros2_yolos_cpp **Creator & Maintainer · 2024 · Active** Production-grade ROS2 nodes for YOLO inference — detection, segmentation, pose, OBB, classification with lifecycle management **Stack:** C++17, ROS2 (Humble / Jazzy), ONNX Runtime, OpenCV, vision_msgs, sensor_msgs **GitHub:** **Detail:** **Markdown:** --- ### motcpp **Creator & Maintainer · 2024 · Active** 10 SOTA multi-object trackers in modern C++17 — 10–100× faster than Python, unified API, ONNX ReID backend **Stack:** C++17, ONNX Runtime, OpenCV, Eigen3, ByteTrack, BoostTrack, GoogleTest **GitHub:** **Detail:** **Markdown:** --- ### SmolVLM2-ROS2 **Creator & Maintainer · 2025 · Active** On-device Vision-Language Model for robotics — SmolVLM2 running via ONNX Runtime inside a ROS2 node for scene understanding and spatial reasoning **Stack:** C++17, Python, ROS2, ONNX Runtime, SmolVLM2, Transformers **GitHub:** **Detail:** **Markdown:** --- ### Depths-CPP **Creator & Maintainer · 2024 · Active** Header-only C++ monocular depth estimation — Depth Anything v2 via ONNX Runtime, real-time on CPU and GPU **Stack:** C++17, ONNX Runtime, OpenCV, Depth Anything v2, CUDA **GitHub:** **Detail:** **Markdown:** --- ### CloudPeek **Creator & Maintainer · 2024** Single-header C++ point cloud viewer — OpenGL 3.3 rendering, arcball camera, no PCL or Open3D required **Stack:** C++17, OpenGL 3.3, GLFW, GLEW **GitHub:** **Detail:** **Markdown:** --- ### dynamic_lidar_interpolation **Creator & Maintainer · 2024** ROS2 C++ package for real-time LiDAR point cloud interpolation — five methods from nearest-neighbour to spline, tested on Velodyne and Ouster **Stack:** C++17, ROS2, PCL, Eigen3, Point Cloud Processing **GitHub:** **Detail:** **Markdown:** --- ### ros2_bag_exporter **Creator & Maintainer · 2024** C++ ROS2 package that exports bag files to images, PCD, IMU, GPS, and CSV — YAML-configured, sqlite3 and MCAP support **Stack:** C++17, ROS2, rosbag2, OpenCV, PCL, YAML-CPP **GitHub:** **Detail:** **Markdown:** --- ### kokoro-onnx-cpp **Creator & Maintainer · 2025 · Active** On-device text-to-speech in C++ — Kokoro TTS running via ONNX Runtime for low-latency voice synthesis on embedded and robot hardware **Stack:** C++17, ONNX Runtime, Kokoro TTS, HiFi-GAN, Audio Processing **GitHub:** **Detail:** **Markdown:** --- ## §02 Research ### FlexiNet: An Adaptive Feature Synthesis Network for Real-Time Ego Vehicle Speed Estimation **Authors:** Abdalrahman Ibrahim, Kyandoghere Kyamakya, Wolfgang Pointner **Venue:** IEEE Access (2025) **DOI:** This paper introduces FlexiNet, a novel adaptive feature synthesis network designed for real-time ego vehicle speed estimation in autonomous driving systems. FlexiNet dynamically synthesizes features across multiple scales and modalities, achieving state-of-the-art accuracy while maintaining the throughput required for real-time deployment on vehicle hardware. --- ### A Graph Attention Network Based System for Robust Analog Circuits' Structure Recognition Involving a Novel Data Augmentation Technique **Authors:** Ali Deeb, Mohamed Salem, Abdalrahman Ibrahim, Joachim Pichler, Sergii Tkachov, Karaj Anjeza, Fadi Al Machot, Kyandoghere Kyamakya **Venue:** IEEE Access (2024) **DOI:** This work presents a comprehensive system using Graph Attention Networks for robust recognition of analog circuit structures. A novel data augmentation technique is introduced to improve generalization across diverse schematic layouts. The system achieves high recognition accuracy on real-world IP analog circuits, enabling automation of a previously manual verification task. --- ### A Comprehensive Generalization of a Graph-Attention-Network GAT Based System Towards Real IP Analog-Mixed-Signal AMS Schematics Structure Recognition **Authors:** Ali Deeb, Mohamed Salem, Abdalrahman Ibrahim, Witesyavwirwa Vianney Kambale, Joachim Pichler, Fadi Al Machot, Kyandoghere Kyamakya **Venue:** 30th IEEE International Conference on Electronics, Circuits and Systems (ICECS) (2023) **DOI:** A comprehensive generalization approach for GAT-based systems applied to real IP analog-mixed-signal (AMS) schematic structure recognition. The work demonstrates that graph attention architectures generalize across schematic styles and technology nodes without retraining from scratch. --- ### A Vertical Systematic Generalization Towards Real IP Analog-Mixed-Signal AMS Schematics Structure Recognition **Authors:** Ali Deeb, Mohamed Salem, Abdalrahman Ibrahim, Witesyavwirwa Vianney Kambale, Joachim Pichler, Fadi Al Machot, Kyandoghere Kyamakya **Venue:** 27th International Conference on Circuits, Systems, Communications and Computers (CSCC) (2023) **DOI:** A systematic approach to generalizing analog-mixed-signal schematic recognition systems across vertical abstraction levels — from device-level schematics to block-level topologies. The method enables recognition at multiple levels of hierarchy without separate training pipelines. --- ### Study of Massive Floating Solar Panels over Lake Nasser **Authors:** Mohamed A. Tolba, Abdalrahman Ibrahim, Ahmed M. El-Nahas, Basem Elboshy, Bassem M. Ramzy, Ahmed M. El-Garhy **Venue:** International Journal of Photoenergy (2021) **DOI:** Comprehensive study on the technical feasibility, energy output modeling, and environmental impact of deploying massive floating solar panel arrays over Lake Nasser, Egypt. The analysis covers thermal performance, water evaporation reduction, and grid integration potential. --- ### Evaluation of Capacitive Deionization Desalination Technology for Irrigation **Authors:** Abdalrahman Ibrahim, O.A. Abdullatef, H.F. Abd-Elhamid, A.M. El-Nahas, M.A. Tolba, Bassem M. Ramzy **Venue:** Desalination and Water Treatment (2020) **DOI:** Evaluation of capacitive deionization (CDI) technology as a sustainable, energy-efficient solution for irrigation water desalination in arid regions. The study benchmarks CDI performance against conventional reverse osmosis and assesses viability for small-scale agricultural applications. --- ## About At AGILOX I lead AI products development. The core platform is a private LLM microservices stack serving 250+ employees — paired with an internal AI gateway that issues virtual API keys for developer and agentic-coding workflows, with routing, fallbacks, and per-team cost dashboards. In November 2025 the gateway handled 12,970 requests and 27.3M tokens at 93.5% success. Commercial API spend dropped ~80% per month, with data staying entirely on-premise. On top of the platform, I run hybrid Graph+Vector RAG for technical documentation and connect LLMs to live ROS 2 telemetry through VAgent using the Model Context Protocol (MCP) for natural-language fleet diagnostics. The inference layer uses vLLM with AWQ quantization on private GPU clusters. I graduated first of class from Misr University for Science and Technology with a B.Sc. in Mechatronics Engineering (CGPA 3.96/4.0, 2018), then spent a year at Infineon Technologies researching Graph Neural Networks for automated analog circuit recognition — improving accuracy 15% and cutting compute time 30% on small-data constraints. An M.Sc. in Autonomous Systems and Robotics at the University of Klagenfurt followed in 2022. I joined AGILOX as Senior Research Engineer, became AI Lead Engineer in May 2023, and was promoted to Head of AI Products Development in January 2026. In parallel I maintain open-source C++ inference tooling built around one constraint: it must drop in without dragging a framework. YOLOs-CPP is the largest — a header-only library running YOLO v5 through v12 on ONNX Runtime and OpenCV, covering detection, segmentation, oriented bounding boxes, and pose estimation, with no Python in the inference path. The strategy-pattern design means switching task types or YOLO versions is a one-line change. The same philosophy carries through ros2_yolos_cpp, Depths-CPP, and the rest of the ecosystem: production-ready inference engines, minimal dependency footprint. --- ## Contact - **Email:** - **GitHub:** - **LinkedIn:** - **ORCID:**