Loïc Busson Email & Phone Number

Fribourg, Switzerland

Villars-Sur-Glâne, Fribourg, Switzerland

👥 Lead paper reviews and knowledge sharing sessions with the team💻 Develop AI models to solve key challenges at Cartier: 📈 Predict future sales and repair orders at different granularities using Time Series Forecasting AI models (such as ARIMA, Holt-Winter, Prophet, NLinear, TiDE, TimesFM) for the supply chain team. I leveraged the python library Darts and studied ways of reconciling forecasts (bottom-up, top-down, NNLS, MinT), as well as returning confidence intervals (quantiles, Monte Carlo dropout). I also designed a Vertex AI Pipeline in Kubeflow to run the pipeline in a serverless manner. 🤖 Develop an internal ChatGPT for the whole Richemont group by leveraging the multimodal Generative AI model from Google (Gemini). I also worked on the Personal Identifiable Information (PII) masking feature to protect employees from leaking their personal information. 📸 Detect serial numbers from pictures using OCR in both individual and batch setups for the factory. I finetuned pretrained models (such as TrOCR) to improve on the Google Vision APIs baseline, while keeping the inference time under 1 second. ⌚ Detect the presence of hallmarks on watches by leveraging the AutoML feature of Google Cloud to train models. I studied several data augmentation techniques to overcome the lack of labeled data, and supervised an intern with annotating images on LabelStudio. 📧 Predict customers' engagement to marketing emails (CatBoost and LSTMs). I also provide feature importance to the marketing team to enhance their understanding of customers' behaviour using embedded methods and SHAP values. 🗂️ Explore RAGs solutions on compliance data to help sales associates find answers to their questions 🌦️ Explore sentiment analysis algorithms to classify comments and extract topics and main feedbacks

Lausanne, Vaud, Switzerland

🥼 Thesis: Local Explainability for Multilayer Perceptron applied to Tabular Data 👨🏻‍🏫 Design state-of-the-art frameworks to provide local and global explainability to Swisscom's models ⬆️ Increase model performance and robustness using the acquired knowledge via interpretability 📝 Write a research paper entitled ”FADE: High-Dimensional Non-Intrusive Feature Attribution for Deep Learning”

Master Of Computer Science, Specialization In Data Science

- 94 credits of courses (3 semesters) - Master thesis at Swisscom (6 months)

Engineer'S Degree

- First year: mathematics, physics, computer science, management and finance courses - Second year: admitted to a dual degree program.

Classe Préparatoire Mpsi/Psi*, Mathematics, Physics, Engineering, Computer Science

- Selective post-High-School scientific courses in Mathematics and Physics leading up to competitive nationwide entrance examinations.

High-School Degree (Baccalauréat) With Highest Honors, Science