Worked on the home page recommender system.
• Built environments to simulate the user/home page interactions and to replay the historical data
• Developed deep learning algorithms, with focus on reinforcement and Bayesian learning
• Computed benchmarks, according to the product constraints (delay reward, position bias, continuous and binary rewards)
• Deployed cloud (GCP) pipeline using managed tools for storage (Bigquery), preprocessing (Apache-Beam / Dataflow), training/serving (tensorflow_gpu / AI Platform)
• Participated paper lecture and research team activities
Developed tools for palm trees detection using machine learning and high-resolution multispectral satellite imagery.
• Acquired and managed the data
• Implemented computer vision solutions (CNN, sliding window, non maximum suppression)
• Experimented at large scale (100 km 2 )
• Visualization and interaction with results (GIS dashboard)
• Presented at FAO/CIHEAM international meeting, Bari (Italy)
Poleon : Real time predictions on transportation demand. https://poleon.co/
Led the design, the development and the maintenance of a software as a service which provided real time predictions to mobility stakeholders.
As data scientist, I managed the five million data collected every day from public and private sources. Including road traffic, Facebook events, meteorology, pollution, RATP, SNCF, Vélib.
• Databases structuring
• Data denoising
• Data engineering
• Predict inflow and demand
• Distribution, visualization and KPI for customers
Developed a prototype for multidimensional data visualization software.
• Automatic generation of pdf reports with interactive visualization (2D, 3D, graphs, shapes, histograms, pie charts)
• Regression methods and detection of outliers
• User interface: display / unveil dimensions, sort visualizations by model / relevance