Achraf Drissi

Client: RATP – Design and automation of ETL pipelines (PySpark, Airflow) to collect and process operational and medical data, with a strong focus on data quality and governance. – Deployment of a real-time Data Lake and Data Warehouse, reducing data access times by 40% and facilitating analysis. – Development of batch processing workflows to manage sensitive records, complying with GDPR and health standards. – Integration of ML models into production via Vertex AI and MLflow, ensuring robustness, reproducibility, and scalability. – Tools: Python, PySpark, GCP (BigQuery, Cloud Storage, Vertex AI, Dataflow, Composer), SQL, Git, Airflow, CI/CD Client: BNP Paribas – Design and deployment of generative AI solutions (RAG, LLM, fine-tuning, Transformers) to automate regulatory compliance processes, reducing manual analysis time by 70%. – Development of end-to-end ML pipelines on Databricks and AWS: data preparation, feature engineering, training, and deployment via Lambda for real-time predictions. – Implementation of MLOps practices to ensure scalability, traceability, and governance of models in production. – Tools: Python, PyTorch, Databricks, AWS (S3, ECS, Lambda, SageMaker), LangChain, SQL, Git, CI/CD, MLflow

Project Title: Distributed Calculation of Similarity Joins Over Large Collections of Tree Structured Data • Conducting an in-depth study of the state of the art in methods for evaluating similarity between hierarchical data. • Development of a new optimised Big Data algorithm, offering a better evaluation (99% improvement) of similarity. • Improvement of the algorithm by integrating the analysis of contextual similarity between files. • Writing and publication of a research paper presenting the new findings in the prestigious IEEE journal.

– 60% improvement in anomaly detection on electrical load time series using the ICS method and LSTM – Development of an imputation method, ensuring business consistency and data quality for decision-making – Predictive modeling using GAM and XGboost, improving forecast accuracy by 33% thanks to the imputation

Project title: Determination of predictive factors for transcutaneous neurostimulation. • Clinical data analysis to identify optimal neurostimulation usage patterns for patients • Patient segmentation based on pain profiles and creation of well-being tracking metrics • Multi-class classification of patient well-being using RandomForest and XGboost + parameter tuning using Optuna • Key factor identification via statistical tests and contribution to certified medical, well being prediction based on the created metrics