I'm Eithan Nakache

Compact ViT with differential attention; up to 96.03% on MNIST and 86.73% on FashionMNIST.

Pre-selected for France’s largest hackathon (Mistral AI x Cerebral Valley). Built a platform for creating task-specific lightweight LLMs using Direct Preference Optimization. Won 3rd prize among 250 teams, competing alongside Hugging Face & NVIDIA.

Winner for car model classification using ResNet18 and explainability techniques (Grad-CAM). Organized by the French Ministry of Armed Forces after 3 months of AI/NLP training, this hackathon marked the start of my passion for competitions.

Developed a privacy-preserving face recognition system using Fully Homomorphic Encryption (FHE). Sensitive data (model weights & embeddings) remains encrypted during computation, with on-device encoding and safeguards against reverse engineering.

Echo is a memory support platform for Alzheimer’s patients and families. It combines a shared digital album, a RAG-powered chatbot for reminiscence, and adaptive memory quizzes. Built during the Mistral AI Hackathon with Alan, alongside 250+ selected participants.