Mohamed Ammar

• Manage multiple data science projects as a project driver collaborating with other data scientists and machine learning engineers. Mainly projects focused on Financial Crime Prevention (FCP).
• Manage communication with multiple internal and external stakeholders (Backend engineers, Product managers, Banking stakeholders, Analysts, External and internal auditors, etc.)
• Collaborate on maintaining a robust auditable documentation for all data science services
• Discuss and drive-forward the OKRs for the data science team and actively contributing in shaping the strategy of the team w.r.t the company's business goals and values
• Participating in the recruitment process within the team and actively strengthening people management skills
• Impact: The designed and maintained models contributed directly to the BaFin (Federal Financial Supervisory Authority) restrictions lift on N26 (i.e. Growth cap lift which allows N26 to onboard as many customers as possible)

• Designed and implemented end-to-end a pre-KYC fraud detection system using a machine learning model (Gradient boosted tree-based model) ; Reducing fraud by ~90% and resulting in multi-million write-offs
• Conduct and manage communication with auditors assigned by BaFin to audit the machine learning model aforementioned involving multiple presentations and walkthroughs
• Design and implement a rule-based model for authentication based on login data using a Kafka infrastructure. At least 10% of the users warned about suspicious login activity took extra steps to protect their accounts.
• Full hands-on implementation of the entire infrastructure of the two models mentioned above including (DB infrastructure, DWH and dashboarding, live monitoring and alerting on DataDog, logging, CI/CD pipeline, credentials management, etc.)

• Analyzed and enhanced N26 chatbot's NLP underlying models resulting in a multi-million (per year) cost reductions related to customer service ops

• Prediction of long-term and short-term time series in the domain of renewable energies
• Tackled supervised and unsupervised learning problematics and implemented solutions based on deep learning and topological optimization
• Designed and implemented reduced order models using linear and non-linear compression techniques
• Designed and implemented image recognition solutions based on convolutional neural networks and an internally written convolutional auto-encoders library
• Accelerated python internal librairies using CUDA and C++ on GPU

Keywords: Deep learning, machine learning, numerical optimization, convolutional neural networks (CNN), time series, feature engineering, data visualization, computer vision, Python, (Keras/Tensorflow, numpy, pandas, scikit-learn,…), MATLAB, C++, Github, Linux