Amine Bennani

Mission DINUM • Python development: developed back‑end for a two django data applications. Applied software craftsmanship, Clean architecture, TDD and pair‑programming best practices for robust solutions. • Orchestration with Celery. ORM & data migrations with SQLAlchemy, and Alembic. • Cloud: deployed application and managed PostgreSQL database on Scalingo for fast deployment and scalibility. • Gitpos: managed version control and deployments using GitLab CI. • Statistical analysis: enhanced data‑driven decision‑making through data analyses for essential services like mask distribution. • Data visualizations: crafted graphs for the applications dashboards with python libraries.

Skills: Data Science · Fraud Detection · Python (Programming Language) · PySpark · Apache Spark · SQL · Cloudera · HiveQL · Bash · makefile · Jupyter · Recommender Systems

Developping Business Solutions, using Data Science & Python programming Language, in a Big Data environment (Cloudera: HDFS, Spark, Hive) • Text matching: contributed to the design of a custom algorithm to consolidate internal sales data sources with recurring panelist referentials, enhancing data integrity and usability, and supporting strategic decision‑making. • Recommender System: contributed to the development of an algorithm mixing Collaborative Filtering and random inputs that strategically suggests the most relevant L’Oréal products for door‑to‑door sales teams, boosting sales efficiency and customer engagement. • Fraud Detection: mentored a trainee in developing an advanced algorithm to identify and prevent revenue loss from unauthorized sales of exclusive L’Oréal professional products. • Python Development: engineered robust data solutions, adhering to TDD and pair‑programming practices to ensure code reliability and maintainability. • Application Maintenance: managed and optimized a Python/PySpark application in a production environment on Cloudera, ensuring seamless operation and performance. Working experience with Hive. • Code Refactoring: transformed legacy code from Jupyter notebooks into a deployable Python package for production deployment. • Scripting & Automation: used Shell scripts & Makefiles for the packaging of Python code, task automation, and the execution of test sce‑ narios.

★★★★★ Machine learning for fraud detection in the context of Display Marketing ★★★★★ ⇨ Context: Studies have shown that ad fraud is likely to represent $50 billion by 2025, and virtually any programmatic buy can be exposed to ad fraud, which is why effective action must be taken. ⇨ Key Accomplishments: • Design of an algorithm for the extraction of features: 6 months of historic web marketing data, 270 GB, 19350 csv files, representing 376 digital campaigns in 31 eurasian countries. • Generating features to simulate the behavior of fraudulent robots. • Desgin of an algorithm for unsupervised learning using density based spatial clustering for multidimentional data; this algorithm detects fraudulent behaviors. • Design of an algorithm for supervised learning using Stochastic Gradient Boosting; this algorithm gives a classification of the placements of digital marketing campaigns between 11 classes representing different levels of infection with fraudulent data (0%, 10%, 20%,..., 100%) • Optimization of the code with profiling and benchmarking: the user time have been reduced drastically to allow the processing time of the features extraction process to go from one week to one day, which is important for automatic daily analysis and fraud detection of web marketing data during the campaigns. ⇨ Gained skills: ✔ Machine Learning: • Feature Engineering • Unsupervised lerning: Clusrtering with density based methods (DBScan) • Supervised learning: Classification (Gradient Boosting Machine) • Bootstrap ✔ R programming: 2000+ lines of code • data.table: data manipulation & transformation • ggplot2 : data visualization • h2o : Distributed Machine Learning • DBscan, LOF : unsupervised clustering • DBI, RPostgreSQL : querying database from R scripts • profvis, microbenchmark : code profiling & optimization ✔ Databases: • SQL: querying a database to retrieve data about the digital campaigns • Postgres

★★★★ Pattern learning for the evaluation of the performance of a Cloud infrastructure ★★★★ ⇨ Context: The performance data generated by IAAS cloud infrastructures contains a mine of information. Using statistical modeling and Queuing Networks, which are particularly adapted to Information systems under the Client-Server paradigm, we can perform quantitative analyses of Cloud infrastructures and services, and manage them in a more efficient and scientific way. ⇨ Key Accomplishments: • Bibliographic research & state of the art. • Modelling of the infrastructure of a SAN (Storage Area Network) using Queueing Networks; Implementation with JMT (Java Modelling Tools) • Inference of the distribution of the inter-arrival times of requests to the SAN using mixtures of Gamma distributions. • Inspiration and opening of intellectual horizons at the contact of very bright and passionate colleagues and a very dynamic company at the forefront of technology ⇨ Gained skills: ✔ Inferential statistics. ✔ Queuing Networks. ✔ R programming: • dplyr: data manipulation & transformation • ggplot2 : data visualization • mixtools: fittinng mixtures of gamma distributions