Curriculum vitae / Applied AI · Computer Vision · Edge AI
Muhammed Ali
Yıldırım
Applied AI and ML engineering. Research, computer vision, desktop software and perception systems on edge devices.
Focus
Applied AI / ML Engineering
From research and model development to inference optimization, desktop tools and edge-device integration.
Python · PyTorch · Computer Vision · TensorRT · DeepStream · PySide6 · Open3D
Education
Gazi University
Electrical & Electronics Engineering · BSc · 2026
GPA 3.58 / 4.00
Contact
Experience
Boğaziçi Savunma Teknolojileri
February–June 2026Computer Vision / Edge AI · Workplace training. From field data to a real-time application: two-stage detection and classification, TensorRT FP16 from 12.8 to 26.0 FPS on a desktop GPU, ByteTrack tracking, distance estimation and interfaces.
TÜBİTAK RUTE
July–September 2025Image processing and deep learning internship. Brake pad detection, ROI, stain and crack segmentation and thickness measurement; Rail3D point-cloud classification with Open3D and LightGBM.
Research & selected work
Drone vs Bird · two-stage real-time detection
A two-stage detection system that tells drones from birds, built during an internship at Boğaziçi Savunma Teknolojileri: YOLO26 detection, YOLO11m-cls classification, a background class, and TensorRT FP16 from 12.8 to 26.0 FPS.
Fixed-wing UAV · detection, tracking and distance estimation
A real-time system for fixed-wing UAVs, built during an internship at Boğaziçi Savunma Teknolojileri: a field dataset, YOLO26L with fine-tuning, ByteTrack identities, a detection console and approximate distance from a single camera.
LabelMate
A local YOLO labeling companion: the model proposes boxes, the person reviews and corrects them. An MIT-licensed desktop application I built on my own in Python / PySide6.
AI-powered search & rescue system
An IMX477 camera on a drone, YOLOv11 on a Jetson Nano and a DeepStream RTSP stream: an end-to-end human-detection system tested in the field. A two-person bachelor’s thesis.
Identifying PCB materials from thermal images · TÜBİTAK 2209-A
A TÜBİTAK 2209-A supported research project and its published article: classifying simulated thermal images of six PCB materials under femtosecond laser irradiation with DualViT-CNN + ArcFace, and choosing energy by material.
Brake pad defect segmentation
From annotation to two-stage segmentation during a TÜBİTAK RUTE internship: brake pad regions, stain and crack masks, and pixel-based thickness output.
Rail3D / Open3D
Exploring HMLS point clouds with Open3D, classification with 3DMASC and LightGBM, and error analysis during a TÜBİTAK RUTE internship.