I am the team captain of our project team for TEKNOFEST, TÜBİTAK, and international competitions.

During my internship at Dpointgroup as an AI Engineer, I worked on several key tasks: Cleaned and pre-processed the data provided by the company, transforming it into a suitable format for further use. Fine-tuned a language model to develop a custom chatbot for the company. Additionally, I created an automation tool using JavaScript, which automated tasks that were previously done manually, significantly improving efficiency.

1. Computer Vision Section: In the computer vision project, precise bounding box calculations were performed using YOLOv10 for mass and calcification detection, and the data was prepared for training. A high-accuracy object detection model was developed using YOLOv10 after two weeks of extensive training, specifically for mass and calcification identification. A CNN-based classification model was designed to predict BI-RADS categories, with accuracy improved through prolonged training. The object detection and classification models were integrated into a threaded system, enabling real-time predictions. The integrated system automatically detects lesions and performs classification on new incoming data, providing real-time inference.2. Natural Language Processing (NLP) Section: Data Preparation and Cleaning: BI-RADS reports were categorized (1-5), with DICOM images converted to PNG. Using Llama 3, text errors were corrected, and additional BI-RADS 1 reports were generated to balance the dataset, followed by Mixtral validation to confirm data quality. Model Development: Logistic Regression and Random Forest models were trained using TF-IDF for vectorization, with GridSearchCV optimizing Random Forest’s hyperparameters. The final model was selected based on accuracy and F1 scores. NER for Medical Entities: With Llama 3 for final text refinement, an NER model was developed using Spacy, achieving high accuracy in identifying anatomical and clinical terms. Evaluation: The optimized Random Forest model reached an F1 score of 0.92, indicating high precision across models.

I worked as a freelancer and managed the setup and maintenance of three different websites on WordPress. My curiosity that started in high school helped me gain experience in this field.