As a Software Engineer at Scale AI, I focus on enhancing the efficiency and accuracy of coding assistant AI models. I evaluate model outputs, identify areas for improvement, and supply high-quality training data to drive continuous model enhancement.

vistrai.com

- Development and testing of computer vision algorithms for challenging illumination and environmental conditions.- Managed deep learning projects from data collection to model training.- Implementing high-performance, GPU-optimized algorithms using CUDA.- Developing image processing algorithms using traditional methods.- Proficient in OpenCV, Keras, TensorFlow, YOLO, MediaPipe, Dlib, RabbitMQ, MongoDB, Flask, Git, Scikit-learn, and SQL/NoSQL databases- Experience with Ubuntu (system administration, shell scripting, performance optimization, creating ubuntu services).- Created and integrated RESTful APIs for seamless backend communication.- Dockerized projects for scalable deployment.- Utilized FFmpeg for live streaming and video compression.- Deployed solutions on Nvidia Jetson, optimizing algorithms for edge computing.--- DMS (Driver Monitoring System) ---- Face Recognition- Pose Detection- Image Classification- Object Detection (e.g., cigarette, mobile phone, seatbelt)- Enhancing images to counter poor illumination conditions- Enabling communication with RabbitMQ messaging- Optimized multi-threaded algorithms, reducing CPU usage by 65%.--- High Security Biometric Verification ---- Live Demo: https://smartir.io/facerecognition- Face Detection- Face Recognition (1:1 and 1:N identity matching) (99.8% accuracy, 100% precision in public benchmarks)- Liveness Detection- Object Tracking- Developing algorithms that achieve high accuracy in poorly illuminated and low-resolution images- Working with depth camera data--- E-Sports Game Analysis (Valorant and CS) ---- Player performance analysis, bullet accuracy, and 2D map visualization.--- Anomaly Detection in Sugarbeet Piles ---- Detecting target regions in thermal images- Calculating the area of the target region from 2D RGB images

As the Data Annotator for the Autonomous Truck Project at Ford Otosan, I was entrusted with the critical task of preparing high-quality, AI-ready datasets from complex sensor data. My responsibilities included:Data Annotation: I meticulously annotated data from cameras and LiDAR sensors installed around the truck. This involved identifying and labeling highways, lanes, vehicles, pedestrians, traffic signs, and various other objects using box, polygon, and line annotations.Point Cloud Processing: I worked extensively with point cloud data collected from LiDAR sensors to ensure accurate representation and annotation of 3D environments.Quality Assurance: I reviewed and verified the annotations done by team members, ensuring the accuracy and consistency of the labeled data.Collaboration: I collaborated closely with engineers and data scientists to optimize the data annotation processes and support the development of robust AI models for autonomous driving.Through these efforts, I contributed to the advancement of Ford's autonomous vehicle technologies, ensuring the precision and reliability of the data used in AI model training and validation.

--- Lane and Drivable Area Segmentation and Object Detection/Classification in Highway Images ---- Developed Deep Learning algorithms for lane and drivable area segmentation, as well as traffic sign detection/classification using data collected and labeled by Ford Otosan.- Utilized technologies such as Python, Pytorch, Tensorflow, YOLOv4, and OpenCV.Key Responsibilities:- Generating masks from JSON file data- Testing the accuracy of generated masks- Creating and managing datasets- Data preprocessing and augmentation- Training segmentation models using SegNet- Training object detection models using YOLOv4- Training image classification models using CNN- Making predictions with test data

--- Drivable Area Detection in Highway Images with Semantic Segmentation ---- Developed a Deep Learning algorithm for drivable area segmentation using data collected and labeled by Ford Otosan.- Utilized technologies such as Python, Pytorch, and OpenCV.Key Responsibilities:-Generating masks from JSON file data to identify pixels representing free space in the images.-Testing the accuracy of the generated masks.-Data preprocessing: Formatting images and masks as features and labels for segmentation, suitable for feeding into the PyTorch segmentation model.-Performing data augmentation.-Training a segmentation model using U-Net.- Making predictions with test data.