العربية

Mohamed Rashad

GenAI Research Lead · Cairo

I’m a researcher and engineer teaching machines to read, hear and speak Arabic. Most of what I make is free and open source, and I love explaining stuff in Arabic 🤍

Experience

  1. GenAI Research Lead, NavidNov 2024 – Present

    Leading research on Arabic generative models: building the evaluation pipeline every model change is measured against, and designing and training large models end to end.

  2. Senior Computer Vision Research Engineer, World RetinaApr 2021 – Jan 2025

    Built recognition-at-scale services as Azure microservices, cutting response times from seconds to milliseconds while keeping compute cost down. Researched object and text extraction from video and text-heavy images, then stayed on as a part-time technical consultant.

  3. Teaching Assistant, Ain Shams UniversitySep 2020 – Jan 2022

    Taught Computational Intelligence and Distributed Systems to engineering undergraduates.

Initiative

Azrak

Azrak is an initiative to make games simply work on Linux. Each project removes one more obstacle between a Linux PC and the games people want to play.

  1. AzrakOSWebsite GitHub

    A PC that works like a console: it boots straight into the launcher, plays with a controller and updates itself, keeping the previous version in the boot menu.

  2. Azrak LauncherWebsite GitHub

    A console-style home screen for Linux: PS5, PS4 and PC games with official artwork, downloaded, installed and played without leaving it.

  3. PES 2021 Online ServerWebsite GitHub

    Brings PES 2021 online play back: quick matches, divisions, co-op and friendly lobbies on a server anyone can host with one command.

Arabic documents & OCR

  1. Arabic-NougatPaper Code Models Demo

    OCR models that turn Arabic book pages into structured Markdown, published with a paper.

  2. Arabic BooksDataset

    8,500 Arabic books, 1.1B tokens, extracted with Arabic-Nougat.

  3. Arabic img2mdDataset

    15,000 document pages paired with their Markdown, built for training Arabic-Nougat.

Arabic speech

  1. Whisper Code-SwitchingModel Dataset Demo

    Whisper fine-tuned for speech that mixes Arabic and English mid-sentence, with the dataset behind it.

  2. Quran RecitationsDataset

    Quranic verses paired with recitations by well-known Qaris.

  3. Text-to-speechEgyptian Arabic Multilingual Orpheus

    Live demos in Egyptian Arabic, in many languages, and with the Orpheus model.

Arabic language

  1. Arabic Tokenizers LeaderboardDemo

    How efficiently each open tokenizer handles Arabic text.

  2. Arabic Auto TashkeelDemo

    Adds diacritics to Arabic text automatically.

  3. Arabic STSDataset

    Arabic sentence pairs with semantic-similarity scores.

Other projects

  1. speak-cliGitHub

    Lightweight text-to-speech from the Linux terminal.

  2. Local Media CenterGitHub

    A self-hosted media center for people who want control over what they watch.

  3. The Simple Arabic ThemeGitHub

    An RTL-first Jekyll theme for writing in Arabic.

Writing & videos

I write in Arabic on my blog about recent AI research: comparisons between models, explanations of training methods, and the best papers I have read. I also publish tutorials on YouTube.