Palaiseau, Île-de-France, France
◦ Conducted an extensive literature review on Transformer-based time-series foundation models and classical stochastic processes (Ornstein–Uhlenbeck, Heston, GARCH) to scope research directions ◦ Designed and coded a scalable Python/PyTorch pipeline for data preprocessing, fine-tuning, and evaluation of foundation models ◦ Prototyped and modified state-of-the-art Transformer architectures from recent papers, embedding domain-specific inductive biases to improve signal extraction in low signal-to-noise settings ◦ Benchmarked candidate models through large-scale back-testing and hyper-parameter sweeps, assessing predictive accuracy, robustness and computational efficiency ◦ Documented findings and presented weekly insights to the Data & AI Lab, contributing to the roadmap for foundation models within Global Markets
Internship Focus: Evaluation and Debugging of Large Language Models (LLMs) ◦ Giskard offers an AI testing platform to help enterprises mitigate AI risks, trusted by clients like Crédit Agricole, BNP, and BPCE ◦ Suggested solutions to clients to improve their ChatBot security and performance ◦ Evaluated and debugged LLM systems, focusing on Retrieval Augmented Generation (RAG) ◦ Led Red Teaming efforts, simulating cybersecurity attacks and recommending architectural improvements to reduce risks ◦ Collaborated with Machine Learning Researchers and Project Managers to align and achieve project objectives ◦ Engaged in continuous research and technological surveillance to contribute to the development of a proprietary ChatBot
◦ Etna Research leverages modern research infrastructure to provide direct indexing solutions for institutional clients across traditional and digital assets ◦ Developed a Logistic Regression Classifier to predict long, short, or neutral positions using the triple barrier method ◦ Engineered meaningful features for the model through extensive testing and analysis ◦ Backtested the strategy across historical data to evaluate the predictive power and refine model parameters for optimal performance
• Experience promoted by University of Pisa • Took care of the manufacturing processes of two components from one of their centrifugal clutches used for gasoline power trowels • Made an accurate analysis of machining and casting processes with the aim of optimizing the production, making use of FEA tools (as SOLIDCast) and CNC programming