experiences, highlight contextual challenges, and collectively troubleshoot common issues
such as ambiguous language, irony etc.
JOSA developed and conducted a training of trainers for Tadwein, and based on that, Tadwein
conducted two two-day training sessions in Cairo in December 2025, targeting CSOs,
researchers, youth-led initiatives and journalists. The trainings aimed to introduce the Nuha AI
model, explain its purpose and functionality, and demonstrate how Egyptian CSOs can use the
tool to detect hate speech and TFGBV on social media platforms. A total of 40 participants
took part in the trainings. The sessions were highly practical and interactive, allowing
participants to test the AI model directly and engage in discussions on its potential
applications, limitations, and areas for further improvement.
The tool can be operated through a standard web browser, without requiring installation of
complex software or access to high-performance computing resources. This simplicity means
that organisations with basic office equipment and moderate internet connectivity will be able
to fully engage with the tool. Importantly, the tool allows users to work at their own pace and
adapt the workflow to suit their operational realities, whether they are conducting continuous
monitoring or analysing specific events, profiles or campaigns.
THE WEBSITE: FOR RESEARCH AND PUBLICATIONS
The website is also an opportunity to showcase organisations in Egypt that address TFGBV in
different ways, including their publications and research on the topic, in both Arabic and
English. The aim is for the website to strengthen links between the Nuha tool and existing
feminist research and advocacy efforts in Egypt.
Adding large scale data to case studies, or conducting smaller studies (focusing on certain
hashtags/ profiles/ time periods/ areas) is now accessible also for smaller organisations. This
means that we hope to see diverse research reports building the case towards prevention and
protection from gendered online hate speech and TFGBV. Also, for those already interested in
the topic, instead of manually sifting through thousands of social media posts, can focus their
efforts on interpreting results, identifying trends, and preparing advocacy materials. It could
also help reveal patterns that have previously been suspected but not proven with data. For
example, analysing trends around elections or other events or incidents.
The trained users will continue to support newer users, and to introduce the tool, both through
campaigns and more hands-on support.
5. CHALLENGES AND LIMITATIONS
The use and misuse of AI currently takes up a lot of band width in public debate. There are
significant challenges as well as risks associated with the use if AI, not least in any kind of
human rights setting. The fact that this tool will only process publicly available information is
important. When using the model, the data harvesting is semi-manual, which limits the risks
relating to automated tracking of online behaviour, which can be related to surveillance,
profiling, or the targeting of vulnerable groups—particularly in regions where digital rights
protection and regulation are weak. Collecting only publicly available data ensures compliance
10