Data Scientist
2022 - PRESENTLevi Strauss & Co.- Led the technical global promotion team, providing recommendations for five global markets for retail stores, outlets, and eCommerce and creating value of around $20 million in additional revenue per year.
- Redesigned the pricing recommendation tool achieving an increase in speed by six times using BigQuery and Apache Airflow.
- Supported the migration from AWS to Google Cloud Platform (GCP), coordinating the work between data engineers and machine learning (ML) engineers.
Technologies: Data Science, Optimization, Pricing, PromotionData Scientist
2022 - 2022Delivery Hero- Designed the first version of the picker scheduling tool that optimized the shifts of the people working in dark stores leveraging Python and mixed-integer programming (MIP), reducing the costs by more than $1 million per year.
- Prototyped the first version of a smart location-based inventory that suggests where to place items optimally to minimize the picking time and other operational activities inside a warehouse.
- Created automated pipelines for autoformatting using Python and SQL codes based on custom rules, helping data scientists to speed up deploys to two hours per person per sprint and to bring uniformity across different teams.
Technologies: Python, Optimization, Data ScienceData Scientist
2017 - 2021Machine Learning Reply- Created text classification to analyze emails automatically, speeding up the entire business process by 10 times.
- Designed an optimization tool for a cruise company that handles embarking and disembarking for more than 50,000 people, resulting in estimated average savings of $1 million per year.
- Conducted 10 lectures for the course "AI and ML: Platforms and vendor solutions" to graduate students enrolled in the second-level master studies in artificial intelligence and cloud at the Polytechnic University of Turin.
- Worked on a recommender system for Reply's internal social network using traditional collaborative filtering methods, item-based models, and NLP techniques. The algorithms handled over 10 thousand active daily users and increased engagement by 10%.
- Redesigned an ML model for swaption prices, achieving better performance by decreasing the mean squared error (MSE) by 10% compared to the previous implementation and lowering the RAM required by 30%, with an increased speed by 1.5 times.
- Built a system to detect and classify various road defects, as predictive maintenance applied to highway asphalt is crucial to cut costs. This object detection model was based on YOLOv5.
Technologies: Python, PyTorch, Deep Learning, Machine Learning, Forecasting, Computer Vision, Natural Language Processing (NLP), Categorization, Regression, Google Cloud Platform (GCP), Unsupervised Learning