Professional experiences
Visiting researcher, cardiovascular disease prediction using polysomnography
Harvard Medical School , Boston
From March 2025 to Today
+ Assembled a multimodal physiological dataset of ~40,000 full-night polysomnography recordings from multiple
cohorts (Human Sleep Project, Sleep-EDF, private studies) using Python, MNE, and HDF5,
+ Pretrained a contrastive “sleep foundation” transformer model in PyTorch, implementing dynamic masking and
padding to accommodate variable channel counts and recording lengths across modalities (ECG, EOG, EMG,
respiratory),
+ Fine-tuned the pretrained model for patient-level classification of cardiovascular disease subtypes (angina, stroke,
congestive heart failure, control), leveraging full-night PSG inputs and achieving an F1 score of 82%,
+ Pioneered end-to-end CVD screening directly from complete-night recordings, eliminating epoch-level
aggregation, and improved predictive accuracy over traditional approaches.
Computer vision intern, deep learning photography style prediction
DXO LABS , Boulogne-billancourt
From August 2024 to January 2025
+ Constructed a synthetic training dataset by applying geometric transformations and center-cropping to raw images
while preserving original color information,
+ Preprocessed photos and extracted deep feature embeddings using Meta’s DinoV2 backbone in PyTorch,
+ Built an attention mechanism and MLP on top of the extracted features to predict photographer-specific white
balance adjustments,
+ Trained and fine-tuned the network end-to-end, achieving high consistency in predicted white balance across diverse
scenes,
+ Assessed correction fidelity using PSNR, achieving an average of 42 dB.