# Soufiane Jadda — Interview context Public positioning: **Data Scientist & AI Engineer**. Questions below are discussion prompts, not claims about undocumented implementation choices. ## Professional background I'm a Data Scientist and AI Engineer with a dual Master's background in Data Science and Decision Support. Through my work in railway inspection and maintenance, I've developed computer vision and AI-based solutions addressing real operational challenges. Beyond industrial applications, my projects explore embedded AI, medical image segmentation, retrieval-augmented generation, and multimodal recommendation systems. ### Public education records - Bachelor's — Computer Engineering, Sidi Mohamed Ben Abdellah University, Fez; 2020 — 2024. - Master 2 — Data Exploration and Decision Support, Dual-degree program with Sorbonne Paris North University; 2025 — 2026. - Master's — Web Intelligence and Data Science, Sidi Mohamed Ben Abdellah University, Fez; 2024 — 2026. Education dates are the supplied study periods; they do not independently establish degree award or graduation status. ## Public experience records ### ONCF Group — Data Scientist Intern Period: 2026-02 — 2026-07. During my Data Scientist internship at ONCF Group, I worked on digitalizing visual inspection in a railway operations context, supporting inspection monitoring, results analysis and operational decision-making. - Translate operational inspection requirements into a data science workflow. - Support inspection monitoring and analysis of inspection results. - Develop application functionality for tracking inspection KPIs. Related project evidence: [AI-Powered Railway Wagon Inspection](/projects/ai-powered-railway-wagon-inspection#experience). ### Moroccan High-Speed Train Maintenance Company (SIANA) — Data Scientist Intern Period: 2025-07 — 2025-08. During my Data Scientist internship at the Moroccan High-Speed Train Maintenance Company (SIANA), I worked on digitalizing railway component identification and spare-parts information workflows in a maintenance and supply-chain context. - Translate maintenance and supply-chain requirements into component identification workflows. - Organize data collection, annotation, preparation and management. - Develop tools for maintaining and expanding component reference information. Related project evidence: [Intelligent Railway Component Identification](/projects/intelligent-railway-component-identification#experience). ## [AI-Powered Railway Wagon Inspection](/projects/ai-powered-railway-wagon-inspection) During my Data Scientist internship at ONCF Group, I worked on digitalizing and automating an industrial visual inspection process. The project focused on detecting residue accumulation on railway wagons through computer vision and real-time video analysis. I developed an AI-based inspection pipeline and integrated it into an application supporting system configuration, inspection monitoring, results analysis, and KPI tracking. The work aimed to improve inspection reliability and support operational decision-making. Documented contributions: Translated operational inspection requirements into an AI-based solution. Developed the real-time video analysis pipeline. Integrated the inspection workflow into an application. Implemented functionality for monitoring results and tracking KPIs. Suggested questions: - How were operational inspection requirements translated into the analysis workflow? - How were video analysis, inspection configuration, monitoring, and results analysis connected? - How were inspection KPIs defined and evaluated? ## [Intelligent Railway Component Identification](/projects/intelligent-railway-component-identification) During my Data Scientist internship at the Moroccan High-Speed Train Maintenance Company, I worked on digitalizing the identification and search process for railway components. The project focused on using computer vision and optical character recognition (OCR) to support component identification and improve access to spare-parts information. I developed an AI-based solution that combined visual recognition with workflows for data collection, annotation, preparation, and management. The solution also included tools for expanding the reference dataset and automatically integrating new component references. The work aimed to reduce identification errors and improve efficiency in maintenance and spare-parts management. Documented contributions: Translated maintenance and supply-chain requirements into an AI-based component identification solution. Developed the computer vision and OCR-based identification workflow. Implemented data collection, annotation, preparation, and management processes. Developed tools for enriching the dataset and integrating new component references. Suggested questions: - How did visual recognition and OCR contribute to component identification? - How were reference data collected, annotated, and prepared? - How did the tools integrate new references and maintain dataset quality? ## [3D Brain Tumor Segmentation](/projects/3d-brain-tumor-segmentation) This project focused on automatic brain tumor segmentation from multimodal magnetic resonance imaging (MRI) scans, within the context of the BraTS Challenge. I designed, trained, and optimized a 3D U-Net model using PyTorch to segment brain tumor regions in three-dimensional medical images. The project explored deep learning for medical image segmentation, with a focus on learning spatial features from volumetric MRI data and evaluating segmentation performance. Documented contributions: Designed and implemented a 3D U-Net segmentation model using PyTorch. Trained the model on multimodal brain MRI scans. Optimized the model for brain tumor segmentation. Evaluated the segmentation results, including the whole-tumor region. Suggested questions: - How were multimodal volumetric MRI inputs prepared for the 3D U-Net? - Which training and optimization choices were compared? - Which evaluation measures and validation procedures were used? ## [Embedded Driving Assistance System](/projects/embedded-driving-assistance-system) This project focused on developing an affordable, real-time driving assistance system adapted to Moroccan roads. The system uses computer vision to recognize traffic signs and communicates relevant information through a multilingual voice assistant. By combining visual recognition with audio feedback, the project aimed to improve traffic-sign awareness while allowing drivers to maintain their attention on the road. The solution was developed for deployment on embedded hardware, with an emphasis on real-time operation and accessibility. Documented contributions: Designed and developed the embedded driving assistance solution. Implemented real-time traffic-sign recognition using computer vision. Integrated multilingual voice guidance for detected traffic signs. Deployed the solution as a low-cost embedded AI application. Suggested questions: - What hardware and recognition approach were selected, and why? - How were recognition and multilingual audio feedback coordinated? - How were real-time operation and driver distraction assessed? ## [Intelligent Hospital Information Assistant for Morocco](/projects/intelligent-hospital-information-assistant-for-morocco) This project focused on making hospital information in Morocco easier to access through a conversational AI assistant. I developed a retrieval-augmented generation (RAG) solution that combines information collection, data organization, and natural-language question answering. The system integrates web scraping and document extraction to collect hospital information, which is organized in a structured SQL database. By combining this information with an LLM-powered question-answering workflow, the assistant provides a conversational interface for accessing hospital-related information. The project explored how generative AI and structured data management can improve access to information distributed across multiple sources. Documented contributions: Designed and developed the RAG-based hospital information assistant. Implemented automated information collection using web scraping and document extraction. Structured and organized collected information in a SQL database. Integrated natural-language question answering using LLMs and retrieval-augmented generation. Suggested questions: - How were collected hospital information and extracted documents organized in SQL? - How did retrieval connect to natural-language question answering? - How were information quality, updates, and answer grounding evaluated? ## [Multimodal Recommendation System](/projects/multimodal-recommendation-system) This project focused on click-through-rate (CTR) prediction within the context of the WWW 2025 MM-CTR Challenge. I developed and optimized a multimodal recommendation system using deep learning techniques, including Transformer-based architectures. The work involved experimenting with different model architectures, evaluating their predictive performance, and refining the approach through offline validation. The project explored how multimodal information can be incorporated into recommendation models to predict user interactions more effectively. Documented contributions: Developed a multimodal deep learning solution for CTR prediction. Experimented with different model architectures, including Transformer-based approaches. Evaluated and compared model performance using offline validation. Optimized the recommendation model based on experimental results. Suggested questions: - Which modalities were combined, and how were they represented? - Which architectures were compared during offline validation? - How were validation splits, evaluation measures, and optimization choices determined? Exact metrics, model variants, datasets beyond the stated contexts, hardware, and deployment environments should only be attributed where explicitly documented. See [Technical Stack](/stack.md) for technology associations.