European Space Agency (The Netherlands)
Project: Lunar Rover Traversability Analysis.
Tech: Conformal Prediction & Geometric Fusion.
Polytechnique Montréal (Canada) - GPA 4/4 Ranked
1st
Theorical and Applied AI, Natural Language Processing, Reinforcement
Learning, Programming, Mathematics.
Mines Saint-Étienne (France) - GPA 4/4 Ranked
1st
Computer Systems, Robotics, Artificial Intelligence, Electronics,
Cryptography, Signal Processing, Languages.
Head of Investment & Quantitative Strategy
Leading a team of 20 analysts managing $140k+ using AI and
quantitative analysis to optimize portfolio.
Designed Transformers and attention mechanisms for uncertainty and risk quantification using Ensemble techniques and Bayesian methods.
Paris, FranceImplemented uncertainty quantification and image segmentation methods to enable rovers to navigate autonomously on Mars under ambiguous terrain.
Noordwijk, NetherlandsResearch on uncertainty and risk quantification for multimodal planetary-terrain segmentation and risk-aware path planning for rover navigation.
Montréal, CanadaEngineering advanced NLP and ML pipelines to extract and analyze financial insights from unstructured data for asset management.
Montréal, Canada
MPSI / MP*
Computer Science
Artificial Intelligence
My academic journey is defined by a dual focus: mastering the theoretical foundations of
mathematics and computer science in France, and applying them to cutting-edge research in
Artificial Intelligence and Robotics in Canada.
Risk-aware path planning through uncertainty quantification, fusing Adaptive Prediction Sets and Kandinsky calibration into spatially coherent risk maps for planetary rover navigation.
T. Gachet, L. Mansilla, D. Magnin · Accepted, SPAICE 2026A unified conformal calibration pipeline for dense multimodal semantic segmentation, applied symmetrically to 2D imagery and 3D LiDAR for reliable planetary rover perception.
T. Gachet, G. Beltrame · Preprint (arXiv) · Target: RA-L / ICRA 2026An epistemic-safe path planner that turns perception uncertainty into a continuous risk field, resolving the freezing-robot problem while minimising worst-case risk exposure.
T. Gachet, G. Beltrame · Preprint (arXiv) · Target: T-RO / ICRA 2026Interactive mineral exploration tool using TensorFlow for real-time rock identification and sustainability mapping (Flask, MongoDB, Streamlit).
Real-time financial platform powered by AWS and LLMs for portfolio optimization, sentiment analysis, economic synthesis, and board profiling.
Development of conversational agents, QA systems, and dependency-based parsing using neural networks and probabilistic models.
Diverse RL projects: from Tabular (MDP, Sarsa) to Deep RL (SAC, TD3, Rainbow). Implementation of Actor-Critic, Policy Gradients, and Model-Based architectures.
Python-based market analysis focusing on portfolio optimization, strategy backtesting, and metrics computation (Sharpe, Alpha, Beta).