Mehdi
ai researcher / engineer — Visionairy. phd in applied mathematics, MAP5/University Paris Cité (2024, oh god).
Day to day my work blends deep learning research and integration with interpretability and model reduction. Tinkering a lot with Dinov3 at the moment.
Outside of that: a cross-platform clipboard sync tool written in Rust, a custom smartwatch built around a Lilygo S3, and a fully local voice assistant on a Raspberry Pi 4 with wake-word detection, STT, tool calling, and TTS.
Generally drawn to image processing, privacy-respecting infrastructure (grapheneOS, linux), and whatever obscure technical problem shows up next (it always does).
