I work on languages and machines for places the tools skip. Lycée Pilote Manouba, Tunisia.
Tunisian Arabic (Derja) NLP
- tunisianmmlu-audit: Is TunisianMMLU Actually Tunisian? A Native-Speaker Audit of a Machine-Translated Dialect Benchmark, with Marek Šuppa (MRL 2026, EMNLP). Two native speakers audited a 400-item sample of the only MMLU-style Derja benchmark; 62% to 86% of it is defective, mostly because the text is Moroccan, not Tunisian. Labels, corrected subset, scanner and evaluation code; data as TunisianMMLU-Verified.
- Tunisian-Derja-NLP-Resources: an inventory of Derja NLP resources, 151 entries, each checked against its source and tagged for access. Zenodo · Hugging Face. Contributions welcome.
Water
- Saqi v2: offline irrigation controller for smallholder farms. One box at the manifold keeps a water account per soil zone and runs the pump itself. First field test on our olive grove: 29% less water than a timer. Saqi v1 is the simulation it grew out of (Moonshot 2026, Top 100).
- Anzar: battery-powered STM32 pipe sensor that tells a leak from normal use and catches taps left running, with no internet. Prototype units run at home and on the grove.
Also: European Girls' Olympiad in Informatics, Team Tunisia (2025, 2026) · Silver, Pan-African Olympiad in Informatics 2026 · four summers teaching informatics to Bac pupils in Manouba, in the summer study sessions my mother runs.
Open to collaboration on low-resource NLP and water tech; open an issue on any repo.