Daouda Thioye

Mentoring and evaluating data scientist, data analyst and ML engineer students

• Built a production-grade MLOps platform on Azure ML: isolated DEV/PROD environments, mandatory PR gating and full model traceability via Git governance. • Set up CI/CD pipelines (Azure DevOps, GitHub Actions) for training and inference of anomaly detection models on electrical equipment images (YOLOx). • Developed PySpark/Spark pipelines on Azure Synapse for ingestion and analysis of Image & LiDAR data (16+ TB, 390,000+ files). • Model deployment and tracking via MLflow (registry, versioning, run comparison). • Power BI dashboards for DataOps/FinOps monitoring and access governance — 50% loading time reduction. • Platform used by 5 industrial partners.

• End-to-end design and deployment of 6 ML models in production (ride-hailing & food delivery): ◦ Automatic dispatch — LightGBM acceptance probability model (<200ms): +5% acceptance rate in large cities, +10%in medium-sized cities, –3 min waiting time. ◦ Dynamic pricing — surge multiplier engine: +30% surge-activated rides during peak hours, +5% revenue per ride. ◦ Driver churn risk scoring (LightGBM / Dataiku): +12% retention on targeted cohort. ◦ RFM segmentation of customers & drivers — integrated into Marketing Looker Studio dashboards. ◦ Food-delivery preparation time prediction (GBM): –15% late deliveries, improved customer satisfaction. ◦ Two personalized recommendation engines (home screen & checkout): +8% AOV measured in A/B test. • Design and deployment of a multi-agent statistical assistant (StatIQ) on GCP enabling non-technical teams to query operational data in natural language (LangGraph, Claude API, BigQuery). • Model industrialization via CI/CD pipelines on GCP (Vertex AI, Databricks, Airflow) with continuous production monitoring. •Led 2 Data/ML squads (10 people: Data Analysts, Data Scientists, Product Managers) — end-to-end delivery, backlog prioritization and team upskilling.

Teaching MSc students how to perform data analysis and visualisation with Tableau : <> process & prepare data with Tableau DataPrep <> create graphs & make analysis with Tableau <> build insightful dashboards and perform a great storytelling

Prepared classroom assignments for Linear Regression

• Built predictive models for sales potential before product launch (NLP on social signals, time series) for fashion retailers: ETAM, Jennyfer, Celio, Decathlon. • Developed a large-scale social media data pipeline (Facebook, Instagram) via ElasticSearch and GCP. • Deployed predictions and visualized results through the internal web application used by client buying teams.

Lecturer in Advanced Machine Learning in M2, Data Science in M1 and Advanced Data analytics in L3.

• PSA/Stellantis (France & Europe) — used vehicle stock forecasting (LSTM), best sales channel assignment model (XGBoost), price/delay decision support, logistics cost reduction. • ArcelorMittal — customer and product segmentation, steel demand modeling, integrated data visualization. • Renault India — social media analytics pipeline (Twitter, YouTube, Facebook), sentiment analysis (fastText), weekly Power BI reporting. Stack: Python, TensorFlow, X

• Modeled 6 customer scores at scale across millions of subscribers: churn, fragility score (6-month departure risk prediction), Orange Bank product propensity, upselling, cross-selling, family composition detection. • Measured internet connection quality dissatisfaction and predicted field interventions via sequence mining (Python).