Bachelor & MSc @Ecole Polytechnique (Paris, France) | M.A. Statistics @UC Berkeley (California, USA)
Bridging the gap between advanced math and high-stakes production. Focusing on building software that works in the physical world. Currently building a platform to automate and centralize health data.
| Category | Key Technologies |
|---|---|
| Core Languages | Python (Advanced), SQL |
| ML/Data Science | PyTorch, Scikit-learn, Pandas |
| Project | Description | Technologies Key to Deployment | Status |
|---|---|---|---|
| Capstone project | Predicting blood-oxygen level across brain voxels via an fMRI as subjects listen to various podcasts | Python, Slurm, HF | Coming soon |
| Personal Project: RL for cybersecurity agent | Built a Harbor RL environment around a deserialization CVE (CWE-502 cat.), to post-train an agent on teacher's trajectories | Python, Docker | Repo |
| Class Project: Pixel classification | Pre-trained a convolutional autoencoder on satellite images of Arctic, classified pixels (cloud VS ice) | Python | Private Repo, original paper |
| Class Project: Traumatic brain injuries classification | Detection of ciTBI for children, helping decide whether they need a CT scan (X-ray) or not | Python | Private Repo, original paper |
| AWS hackathon | Built an agent to check for potential leakages of data/sensitive information coming from other agents' actions. | Python | Repo |
| Class Project: Genetic Algorithm | Developed a genetic algorithm to select the best features of any dataset, in order to predict the values of a target feature with regression methods. | Python | Private Repo |
| Research Project: Membership Inference Security | Analyzed the promise of quantization of models’ weights for improving Membership Inference Security (MIS). Developed an algorithm that improved the average security metric by a factor of 10,000 compared to an equivalent SotA quantizer. | Python (PyTorch), Slurm | Private Repo, original paper |
| Research Project: Anomaly Detection for Images | Developed and trained Convolutional Auto-Encoders to detect abnormal images from an industry-based dataset (MVTec). Reached an AUC of 0.95, significantly outperforming PCA and linear auto-encoders. | Python (PyTorch) | Paper |
| Class Project: Sub-event detection in Twitter streams | Detected events occuring during a soccer game thanks to Twitter data. Processed text through different embeddings, selected relevant features and compared models performances. | Python (Pandas, Scikit, NLTK) | Paper |
| Research Project: Multifractal Modeling in Finance | Developed an algorithm to forecast market volatility and price options. Improved the Markov-Switching Multifractal model (MSM) to better fit complex financial returns distribution and enhance forecasting accuracy. | Python (Pandas, SciPy) | Repo, reference |
Due to the sensitive nature of academic research, a portion of my production-grade code remains private.
I am highly enthusiastic about discussing these projects and the architectural choices.