In a rapidly digitizing world, automating identity document processing has become increasingly important in areas such as e-governance, banking, border control, and many other fields. Therefore, identifying and extracting data with high speed and accuracy is crucial. A significant challenge for developing machine learning models is the scarcity of publicly available data and documents, especially for non-English documents such as Russian internal passports, due to privacy concerns. Therefore, the optimal solution is to generate synthetic data that mimics the original documents but uses fake data. In this study, an integrated hybrid approach is presented to identify and extract information from the generated Russian passport dataset. Deep learning and optical character recognition (OCR) techniques are combined. A dataset of over 1,000 passport images was first created based on a realistic passport template. This template was fed with synthetic personal data generated from the Faker library. Each passport includes identity attributes such as last name, first name, gender, date of birth, issue details, and MRZ lines. Faces are generated using gender-matched synthetic images, and all fields are displayed in Russian. The process of detecting and extracting data from the generated documents involves two main stages. First, we apply the YOLOv8 object detection model to locate key text fields. In this study, we compare the performance of YOLOv8n (Nano) and YOLOv8m (Medium) to evaluate performance differences. Second, all detected regions from the first stage are passed through Tesseract OCR, optimized with support for multiple languages, including English and Russian. The outputs are then structured into clear JSON records for subsequent analysis. Our evaluation, based on benchmarks including precision, recall, mAP@0.5, and mAP@0.5:0.95, reveals that the YOLOv8m variant significantly outperforms YOLOv8n. YOLOv8m achieved a precision of 0.999, a recall of 1.000, 0.995 mAP@0.5, and 0.931 mAP@0.5:0.95, demonstrating its efficiency in detecting small or vertically stacked fields such as region code, year, and gender. Further- more, MRZ detection improved significantly when switching to the intermediate model. These results confirm the feasibility of using synthetic datasets for document analysis tasks and highlight the effectiveness of the hybrid detection and OCR pipeline.