Mustapha Zakaria MERABET

- Designed and optimized batch and streaming data pipelines with Spark (Scala & PySpark) on Azure Databricks. - Developed ETL/ELT pipelines for ingesting and transforming data from PostgreSQL, CSV/XML, and REST APIs using Apache NiFi, dbt and Talend. - Modeled data in Medallion architecture (Bronze/Silver/Gold), optimized Delta Lake with partitioning, Z-Ordering, and compaction. - Enabled schema evolution, time-travel, and streaming/batch consistency using Delta Lake. - Implemented data governance with Unity Catalog: access controls, cataloging, and auditing. - Created CI/CD pipelines with: Unit tests (PyTest, ScalaTest), Docker-based deployments, GitHub Actions & Jenkins for automation to Databricks Jobs and Azure Functions - Built serverless data-processing apps using Azure Functions, triggered by Blob Storage, HTTP, or queues. - Exposed Gold-layer data to AWS S3 and Redshift for BI teams. - Processed, stored, and optimized enriched transcription data in BigQuery for analytics. - Implemented data quality checks, anomaly detection, retention policies, and GDPR compliance. - Authored detailed technical specs and solution architectures, participated in Agile/Scrum ceremonies. - Implemented RAG (Retrieval-Augmented Generation) architectures. - Developed OCR pipelines using Tesseract, Docling, LayoutLM, and pdfplumber for semi-structured and unstructured data extraction from PDFs. - Applied regex, spaCy, and NLTK for text segmentation, NER, and normalization, output structured in JSON or SQL. - Built intelligent agents and chatbots using LangChain and LangGraph. - Set up Kubeflow pipelines for automated model retraining (Continuous Training). - Deployed Evidently AI for Continuous Monitoring of model accuracy and data drift detection. - Used MLflow for experiment tracking and model versioning. - Provisioned and managed cloud infrastructure using Terraform, enabling automated deployment and scaling of data and ML services.

- Developed and maintained cross-platform mobile applications for clients across various countries and industries, including e-commerce, healthcare, education, and social networking. - Delivered custom solutions tailored to each client's business needs, adapting to different technical stacks and project scopes. - Utilized diverse state management approaches such as Riverpod, Bloc, and GetX, depending on project requirements and scalability goals. - Integrated a variety of databases and backends (e.g., Firebase, Supabase, REST APIs, GraphQL) to support real-time data, authentication, and cloud storage. - Employed popular Flutter packages and tools to enhance app functionality and user experience, with a focus on performance, maintainability, and responsiveness. - Built scalable applications designed to handle high volumes of users, with a focus on performance optimization and stability. - Collaborated with international teams and stakeholders through all phases of the development lifecycle, from planning to deployment and post-launch support.

• Built a cross-platform mobile app (Android & iOS) integrating multiple on-device ML models. • Extract meaningful data and insights from heterogeneous datasets. • Performed model selection under strict constraints, balancing accuracy, speed, and privacy. • Develop and implement innovative methods to effectively categorize and group data. • Applied clustering algorithms (DBSCAN, Chinese Whispers) to group images by themes and visual similarity. • Developed an automated tagging pipeline using CLIP for keyword/caption generation. • Fine-tuned detection models for task-specific performance and optimized on-device deployment on smartphones. • Integrated models with Core ML for accelerated performance on iOS devices. • Applied NER and time/place metadata extraction for fine-grained semantic search. • Enabled semantic, text-based image search through NLP techniques. • Collaborated with design teams to incorporate UX heuristics into algorithmic decisions. • Evaluate and select optimal models for various tasks, balancing both performance and efficiency. • Reduced inference time from 380ms to 150ms using float16 quantization and deployment via Tract (ONNX/NNEF). • Achieved 95% smart search accuracy, validated through internal testing. • Used technologies: Python, C++, Pytorch, Keras, Tensorflow Lite, Core ML, Scikit-Learn, Hugging Face, SpaCy, OpenCV, NumPy, Seaborn, Pandas, ObjectBox , Git.

• Build an intelligent system that seeks to estimate crowd size in real time in a crowded scene. • Study of different existing real-time crowd counting models specially for edge devices and identification of the model to exploit (YOLO, R-CNN, Faster R-CNN and SPP-Net). • Developing YOLO-CROWD, a lightweight crowd counting and face detection model that is based on Yolov5s. • Fixing the problems of face occlusion, varying face scales, and other challenges of crowd counting in a way that does not compromise the system’s performance in favor of its size. • Creating our own dataset using Roboflow, in order to customize the training, testing and validation sets and getting the best results. • Used Libraries and Tools: Pytorch, Tensorflow, Numpy, Pandas, OpenCV, CUDA, TensorRT, NVIDIA® Jetson Nano Developpement Kit, Raspberry Pi Camera Module 2, Git.

Within this company I worked on a mobile application called Dowaya. - Dowaya is a cross-platform mobile app for finding nearby doctors and pharmacists, with search options by name, specialty, location, and other filters. - I integrated features for users to get directions to healthcare professionals or contact them directly, while enabling doctors and pharmacists to manage their profiles and availability. - I implemented a dashboard for healthcare professionals to track usage statistics and update their profiles, including setting availability (permanent or absence days).

- Coding and updating applications or software packages to handle specific tasks. - Reviewing and implementing solutions for bugs and enhancements within software. - Applying agile methodologies, facilitating collaboration with stakeholders and continuous improvement. - Performing revision and repair of existing programs, increasing operating efficiency or adapting to new requirements. - Used technologies: Laravel, Flutter, Agile, Microservices, Docker, Bootstrap, Java, Dart, Php, JavaScript, SQL, NoSQL Databases, Git.

As part of my studies at the Higher School of Computer Science in Sidi Bel Abbés, I completed an internship at the Hasnaoui company. This internship was an opportunity for me to understand the proper functioning of a company using IT tools. It allowed me to put into practice the theoretical and methodological knowledge acquired during my training and to enrich my knowledge.