Baptiste Gerbouin

• • Designed a scalable spatio-temporal database (>1.5 TB, PostgreSQL) integrating heterogeneous data streams.

• • Automated ETL pipelines and optimized queries: reduced extraction time from 24h to 20 min.

• • Developed hybrid models (GR5J + Random Forest) and Feature Engineering on high-frequency data to maximize NSE score.

• • Designed a decision support tool (Interactive Dashboard) and provided pedagogical explanation of results to a multidisciplinary team (Hydrogeologists).

• • Analyzed complex data (85 million customer/product lines) to develop a predictive statistical model.

• • Optimized customer recommendations using deep learning (LSTM) and machine learning (LightGBM) methods.

• • Evaluated performance with the Hit Rate @10 metric, generating a 7% increase in repeat purchases.

• Designed an image classification pipeline for automated wind turbine detection using PyTorch.

• Trained and compared architectures: CNN (From Scratch) vs Transfer Learning (ResNet).

• Implemented Data Augmentation and pre-processing strategies to improve model robustness.

• Optimized hyperparameters and analyzed performance metrics to validate generalization.