MSc in Data Science & Machine Learning @ Reichman University
Building and deploying high-performance models with a focus on efficiency (ONNX/TensorRT) and MLOps.
🏆 1st Place - Kan News Hebrew Synthetic Voice Competition
🐦 Top 2% (38/2025) - Kaggle BirdCLEF 2025 Competition
|
An end-to-end Text-to-Speech engine designed for expressive prosody, robust zero-shot voice cloning, and production-grade latency. The entire stack is optimized with ONNX & TensorRT for efficient, private deployment. ➡️ View Live Demo & Details |
A practical Niqqud (vowelization) model that incorporates phonetic awareness to achieve superior accuracy on modern Hebrew. This work emphasizes sentence-level evaluation and clean, deployable code. ➡️ Read the Paper / Project Page |
|
A level-aware 2.5D/3D classification pipeline for analyzing lumbar spine MRIs (Sagittal T1/T2, Axial T2). The model uses segmentation-guided cropping and a transformer-based combiner to achieve robust stenosis grading across L1–L5 vertebrae. Kaggle Competition Entry |
Developed a highly efficient pipeline for large-scale audio classification. Achieved top 2% (rank 38/2025) with an AUC of 0.902 through disciplined validation, weighted-blend ensembling, and optimized audio feature extraction. Kaggle Competition Entry |
