At Dataguess, I built and deployed computer vision systems for industrial clients, including object detection, anomaly detection, tracking, and counting. I worked across the model lifecycle, from data collection and labeling through training, evaluation, and refinement.
I deployed and optimized models on NVIDIA Jetson and Luxonis OAK devices using TensorRT and ONNX. I also delivered poka-yoke, counting, and tracking projects through site visits, requirements gathering, and customer handover.
At Yolda, I developed Python ML and optimization pipelines for vehicle-to-trip matching and fleet utilization. I built data and reporting pipelines with MongoDB and AWS, and contributed to workflows associated with an approximately 10% reduction in transportation costs.
Earlier, at PEAKUP, I built a vector search pipeline with Azure Cognitive Search to ground a GPT model on company data, and reduced the vector index size by a factor of five. My academic work includes a CNN for indoor object recognition and an RNN for Turkish Sign Language video-to-text translation, which achieved 89.92% accuracy.
