Final-year Data Science & Machine Learning student at ENSAE Paris, part of Institut Polytechnique de Paris .
Currently double degree student at ENSAE Paris in the DSSA 3A track (Data Science, Statistics and Learning) and the M2DS Research Master in Applied Mathematics and Computer Science at École Polytechnique (M2DS program).

I have worked on professional projects ranging from CNN/Vision Transformer models (IDEMIA, 2 patent disclosures) to forecasting models for private investments (Quantilia) and large-scale analytics (Canal+ International).
Beyond these roles, I develop numerous ML, deep learning, and quantitative projects showcased below.

I am seeking research-oriented opportunities in ML/AI, quantitative modeling, causality, time series, and large-scale applied research.

Contact:

• Email:
  • maxime[dot]coppa[at]polytechnique[dot]edu
  • maxime[dot]coppa[at]gmail[dot]com
• LinkedIn: linkedin.com/in/maxime-coppa

• CAUSALITY
  Exploring the fascinating realm of causality as a research intern at Capital Fund Management (CFM).

A selection of some projects.

• Low-Rank Optimal Transport in OTT-JAX
  Contributing to the open-source OTT-JAX toolbox by implementing a low-rank optimal transport solver based on factor relaxation and flexible latent couplings. Developed modular components for Sinkhorn projections, factor updates, latent coupling optimization, and visualization; validated the method on synthetic transport experiments. Supervised by Prof. Marco Cuturi (ENSAE Paris, Apple).
  Project

• QRT Asset Allocation Data Challenge
  Forecasted daily return signs across 65 asset allocations under low signal-to-noise conditions. Designed leakage-free validation and residual modeling strategies combining market-level and allocation-specific signals.
  Project

• Molecular Graph Captioning
  Developed a graph-to-text framework generating human-readable molecular captions from graph representations. Combined graph neural networks, pretrained language models, contrastive learning, and retrieval-based methods as part of the ALTEGRAD course within the MVA / M2DS program.
  Project

• Object Verification (PyTorch, Transformers, OpenCV) Deep learning models verifying whether two images represent the same object from different views and unseen classes.
  Project

• ML for Portfolio Management and Trading Stock prediction using US Federal Reserve (FRED) data, supervised by Pr. Sylvain Champonnois (CFM).
  Project | Course

Some personal and academic projects I open-sourced on GitHub.

• Time-Series Foundation Models for Classification
  Adapted pretrained Moirai encoder representations to LSST time-series classification, evaluating pooling strategies, attention heads, patch scales, mask-token usage, and fine-tuning regimes including LoRA and full fine-tuning. Supervised by Prof. Romain Tavenard (IRISA Team).
  Project

• Low-Rank Optimal Transport in OTT-JAX
  Implemented a low-rank optimal transport solver with factor relaxation and flexible latent couplings using OTT-JAX; developed modular components for Sinkhorn projections, factor updates, latent coupling optimization, and visualization. Supervised by Prof. Marco Cuturi (ENSAE Paris, Apple).
  Project

• Molecular Graph Captioning
  Developed a graph-to-text model for molecular caption generation using graph neural networks, pretrained language models, contrastive learning, and retrieval. Supervised by Prof. Prof. Michalis Vazirgiannis (LIX / École Polytechnique).
  Project

• Object Verification
  Developed deep learning models to verify whether two images represent the same object from different views and unseen classes, using PyTorch, Transformers, and OpenCV.
  Project

• Reinforcement Learning: Model-Agnostic Meta-Learning
  Investigated fast adaptation with MAML-REINFORCE, benchmarked against REINFORCE and Q-learning, and extended experiments to regression and sinusoidal function fitting. Supervised by Prof Luiz F. O. Chamon (CMAP / École Polytechnique)
  Project

• MLOps Prediction Market
  Built a production-oriented LightGBM pipeline for Kalshi binary event markets, including feature engineering, Optuna tuning, walk-forward backtesting, MLflow tracking, FastAPI serving, Docker containerization, CI/CD, GitOps deployment, and a Quarto website. Supervised by Prof. Romain Avouac and Lino Galiana (Insee).
  Project | Website | API Docs

• QRT Asset Allocation Data Challenge
  Forecasted daily return signs across 65 asset allocations, addressing low signal-to-noise conditions through leakage-free validation and residual modeling.
  Project

• ML for Portfolio Management and Trading
  Built and evaluated stock return prediction models using U.S. FRED macroeconomic and sentiment data, Markowitz optimization, and LightGBM regressors. Supervised by Prof. Sylvain Champonnois (CFM).
  Project | Course

• Pricing Financial Options
  Implemented Black-Scholes, binomial tree, and Monte Carlo methods in C++.
  Project | Report

• Advanced Machine Learning
  Implemented Transformer architectures for stock return prediction on the Jane Street dataset and benchmarked them against gradient boosting models. Supervised by Prof. Austin Stromme (CREST).
  Project

• Myeloid Leukemia Survival Prediction
  Built survival models on clinical and genomic data, evaluated with the IPCW C-index.
  Project | Competition

• Canal+ Loyalty Program Analysis
  Estimated and predicted loyalty program performance in Africa using data from more than 500K subscribers.
  Project

• Linear Time Series Modeling
  Modeled and forecasted industrial production indices using ARIMA models in R.
  Project | Report

Programming: Python, R, C++, Bash, LaTeX
Machine Learning: PyTorch, TensorFlow, Scikit-learn, LightGBM, XGBoost, OpenCV
Data & Scientific Computing: NumPy, Pandas, Polars, SQL, JAX, OTT-JAX
MLOps & Tools: Git, Docker, FastAPI, MLflow, Optuna, CI/CD, Quarto, uv