Othmane El Omri

AI / ML • Lead of AI article translation project aligned with Le Monde In English edito- rial guidelines, accelerating 3x the publication process. – Developed from scratch an AI translation system in collaboration with Le Monde journalists. – Implemented automated evaluation for translation quality and respect of Le Monde style rules. – Industrialized the solution as an API (FastAPI) and integrated it into Le Monde’s CMS (content management system). • Journalists AI assistant for automated article generation and content recom- mendation. – Development and improvement of an AI-driven article generation system based on AFP (Agence France Presse) dispatches, including automated recommendations for relevant articles, tags, and images. • Built ML models for churn prediction and subscription propensity. Data Engineering / Architecture • Lead of migration and architeture of the journalists’ AI assistant from Azure to GCP, including data storage, queues, cloud functions, API, vector store, etc. which reduced the cloud costs by 46% and latency by 80%. • Developed and maintained main data pipelines on GCP (Events ingestion, most read articles, top next articles, etc.), handling both streaming (Pub/Sub, Dataflow) and batch processing (Dataform/BigQuery).

- Deployment of various customer segmentations and scores of Carrefour clients via Airflow DAGs. - Implementation of MLOps end-to-end workflow on GCP with Vertex AI (Custom container training, batch prediction, evaluation, monitoring, explainability etc.)

- Modeling of the impact of marketing investments (TV, radio, OOH, press, etc.) on brand awareness and brand consideration. - Modeling of 3 sectors: insurance, tourism, and e-commerce. - Use of pooled models with clustering and PCA.

Projects : - Drift control: Anomaly detection in engine testbeds - Condition based maintenance: monitoring the conformity of two types of machines - filling machines and machining centers - in several Renault plants around the world - Usage based maintenance: wear prediction of brake pads and tires based on the driving behavior - Road health: visualization of hazardous areas based on Renault driving data Misc : - Leading the framing of several business needs and data evaluation - Developed demos with Intel Movidius VPU and a Raspberry to show the ML capabilities to several businesses - Given data trainings "Python for Data Analysis" to business Tech stack : Python (sklearn, tensorflow, xgboost etc.), Google Cloud Platform (Vertex AI ,BigQuery, etc.) Others : Kubeflow, Spark, Folium

Taking part in three projects dealing with security of Intelligent Transportation Systems (ITS) Cybersecurity for Intelligent Transport (CTI) • Development of a solution for clustering/classification of drivers based on their driving style while maintaining their privacy for insurance applications. (Privacy Preserving k-means, Secure Multi-party Computation, Time Series Clustering) • Development of an attack on the vehicular CAN bus, visualization of the impact on a driving simulator. Secure Cooperative Autonomous Systems/ITS Security (SCA / ISE) • Deep learning for classification of types of misbehaviour in a network of connected vehicles. (Misbehavior Detection, RNN, LSTM) • Participating to ETSI (European) standardization of vehicular communications (documents TS 102 941 and TS 103 097). • Clustering of ITS use cases for security analysis. • State-of-the-art, analysis and specification of solutions for several research topics: - Privacy mechanisms for ITS (pseudonym certificates) - Architecture and protocol of an Vehicular Ad-hoc Network PKI Industrial partners : Renault, PSA Peugeot Citroën, Valeo, IDnomic, Airbus, Alstom, etc

Objective: performing biometric operations while preserving the privacy of biometric data (fingerprints, iris) by using Secure Multiparty Computation techniques. State-of-the-art (Yao, GMW, Oblivious Transfer, malicious attacker case), specification of new protocols, C++ implementation. Results published in an international conference : Boosting GSHADE capabilities: New applications and security in malicious setting. SACMAT, 2016.

Protection of sensitive data in a printing system.

Cryptographic analysis of the Bitcoin blockchain and protocol.