Automated KYC @Dyos

Between 2019 and 2021, I held the position of Lead Machine Learning Scientist at Dyos Technology GmbH, later known as AICOR Verwaltungs GmbH. My role involved leading a multidisciplinary team of scientists, machine learning engineers, and developers in creating Eli-Ident, an automated KYC product. Our objective was to create a product that enables “strong customer authentication” through video for financial institutions like payment providers or banks, as specified by BaFin. This product was developed exclusively using proprietary algorithms and ML-based models and evolved into qundo.de.

Challenges

The primary challenge in developing this product was maintaining a near-zero false positive rate without compromising the user experience, striking a delicate balance between security and user friction. Additionally, the project demanded the training of multiple models for different tasks under the constraint of limited data access, particularly for large datasets of ID cards.

Tasks and contribution

My leadership in the project encompassed various aspects of backend development, from experimental design and algorithm training to database management and quality control. Key contributions included:

• Object Detection: Developed a document segmentation model for unconstrained environments with over 99% binary accuracy on a quality-controlled test set. This involved curating a proprietary video database with various ID documents and training a deep learning model.
• OCR: Invented an algorithm for extracting text from ID-type documents in unconstrained settings—a significant challenge at the time. This algorithm integrated multiple steps and sub-models to bypass the high costs of direct training.
• Face Recognition: Achieved greater than 99.9% accuracy on the LFW benchmark by creating a comprehensive face image database for identity verification, allowing a highly reliable face-matching model.
• ID Security Features: Developed algorithms to verify the authenticity of German ID card security features, such as holograms and guilloche patterns, using a mix of deep learning and computer vision techniques.
• Liveness Detection: Formulated algorithms capable of discerning the authenticity of documents and the identity of individuals, enhancing security against fraudulent activities.