I'm Eithan Nakache

Platform for task-specific lightweight LLMs with preference optimization; ranked Top 3 among 250+ teams.

Car-model classification with ResNet18 + explainability (Grad-CAM); end-to-end data to deployment prototype.

Privacy-preserving face recognition using FHE (Concrete-ML); embeddings et poids chiffrés pendant l’inférence.

Memory support app for Alzheimer’s: shared album, RAG chatbot, adaptive memory quizzes.

Compact ViT with differential attention; up to 96.03 percent (MNIST) and 86.73 percent (FashionMNIST).

Reimplementation and study of CCVAE and semi-supervised VAEs on MNIST and FashionMNIST; how VAEs capture label structure and leverage partially labeled data.