UAE
AI without internet: How graduate from UAE's MBZUAI creates smart apps that work offline
Daniel Gebre comes from a small city in Eritrea, where internet access was scarce, prompting him to think of how technology could be made to work in such environments
Fri 8 Aug 2025
By: Haneen Dajani
Photo: Daniel Gebre/MBZUAI
When
Daniel Gebre was growing up in Dekemhare, a small city in Eritrea, internet access was a rare luxury. He and his classmates would take turns using a single campus connection, share downloaded files, and rely on offline copies of
Wikipedia just to study.
That experience stayed with him — not as frustration, but as fuel.
Growing up in a place where internet access was scarce, I know the frustration of being curious and eager to learn but constantly being held back by limited digital resources,
he said.
That challenge pushed me to think about how technology could be made to work in environments like the one I came from.
Now 27, Gebre is a recent graduate of Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), where he focused his master’s research on how to bring artificial intelligence (AI) tools to people without internet access.
His thesis project,
iShrink, compresses large language models (LLMs) so they can run offline on mobile devices — an innovation aimed at students and professionals in low-connectivity areas.
There were times in Eritrea when even downloading a simple document or video was a challenge,
he said.
iShrink is about making sure people do not have to wait for internet access to be part of the AI conversation.
How it works
In simple terms, iShrink is a framework designed to make large language models smaller and more efficient without significantly affecting their performance,
Gebre explained.
It does this by identifying and removing parts of the model that are less important, then fine-tuning the remaining components so they still work well. This makes the models faster, lighter, and easier to run on devices with limited resources.
Shrinking smaller models poses a unique challenge, he said, because they contain fewer redundant parameters compared to larger models.
Despite this, we achieved about a 22.5 per cent and 19.7 per cent size reduction on models such as LLaMA 3.1-1B and Falcon 1B respectively.
iShrink currently supports LLaMA, Falcon, and Qwen models, with plans to expand to other open-source architectures in the future.
To test the tool, Gebre developed a mobile application that runs entirely on a local device.
The results were promising,
he said.
In the future, I aim to enhance it with voice and multimodal capabilities to make it even more useful in real-world scenarios,
he added,
I plan to make iShrink fully open source so it can be used and improved by the AI community.
Scholarship to the UAE
Gebre moved to the UAE in 2019 on a scholarship from the Ministry of Education, awarded to top engineering students in Eritrea. He completed his undergraduate degree in information technology at Zayed University with a focus on cybersecurity. It was during a research internship at MBZUAI that his direction changed.
Mentored by Dr
Moayad Aloqaily and Professor
Mohsen Guizani, he became immersed in the possibilities of AI and how it could address real-world problems.
They really encouraged me to explore AI more seriously,
he said.
I had planned to pursue my master’s elsewhere, but Professor Mohsen kept urging me to apply to MBZUAI. Eventually I did, and that decision changed everything.
Gebre described his first semester of graduate school as the most difficult academic experience of his life. He spent long nights reinforcing his math skills and adjusting to the pace of the programme.
I had to catch up fast,
he said.
There were days I studied until three in the morning just to stay afloat.
His perseverance paid off. He completed his thesis, co-authored three academic papers, and presented at international conferences, including the IEEE International Conference on Human-Machine Systems in Abu Dhabi.
You really grow when you take ownership of your learning,
he said.
It was difficult, but transformative.
From research to real-world impact
Since graduating in May 2025, Gebre has joined
Inception, a G42 company, as an Applied Scientist. He is now working on domain-specific AI solutions for industry — a transition supported by a prior internship at the Technology Innovation Institute.
That was the first time I saw what it takes to bring AI into production,
he said.
It gave me a much clearer picture of how research turns into impact.
While his focus today is on refining his skills in industry, Gebre’s long-term mission remains unchanged: to expand access to technology for underserved communities.
In Eritrea, many students still have not heard of tools like ChatGPT or Gemini,
he said.
Even at college, they often have to gather around one internet access point to download materials. The gap is real, which means the potential is even greater.
Future plans
Absolutely,
he said, when asked whether he plans to return to Eritrea to work on digital infrastructure and education.
Given the current technology gap in our country, I want to be among the pioneers who lay the foundations for a thriving tech industry in Eritrea. My goal is to lead AI enablement programs across various domains and initiatives.
When it comes to Eritrea’s many ethnic groups and languages, he believes the lack of digital representation is an issue that can and should be addressed.
Our population is around six million, and while English is used in some areas, it is not universal,
he said.
The main challenges are data and computation. As a small population country, gathering sufficient data is not easy. Most of our literature in local languages exists in hard copy form, so collecting these materials, digitising them, and preparing them for training is a significant but achievable task. With adequate resources and collaboration, it is certainly possible to build such a model.
At the moment, I am fully engaged in my role at Inception. In the future, I plan to launch an initiative to bring together individuals with backgrounds in AI, machine learning, and natural language processing to contribute to open-source projects that develop models aligned with Eritrea’s diverse languages, cultures, and values.
Advice to students from under-served communities
During a panel discussion at MBZUAI, one of my professors said that the key to excellence in any field is mastering the fundamentals, no shortcuts. That advice completely changed my perspective, and I would pass it on to anyone starting out in AI,
he explained.
For students from backgrounds similar to mine, I would add that talent and good academic performance are not enough. The right mindset, consistent effort, strong interpersonal skills, and building a solid professional network are equally important in determining how far you can go.
When he received his degree, his mother was in the audience — a proud moment marking how far he had come.
I am so grateful to the UAE for giving me this opportunity,
he said.
It changed the course of my life.
He hopes his work is only the beginning.
If iShrink or any part of what I have done can help students like me learn, explore, and grow — no matter where they live or what resources they have — that’s the kind of impact I want to keep building.
China, 
Russia, 
North Korea, 
Eritrea, 
Burkina Faso.
