Mahmoud Alyosify

Names

Alyosify — also written Al-Yousify — is the family name by which he is known professionally; in Arabic it is written اليوسفي. His full legal name is Mahmoud Sayed Youssef Kotb (محمود سيد يوسف قطب), and he also appears in records and course listings as Mahmoud Sayed Youssef (محمود سيد يوسف). All of these names refer to the same person. His ORCID record lists his legal name, with Mahmoud Alyosify as the name he publishes under.^[6]

Mahmoud AI is not another name for him: it is the conversational assistant on his website, which answers questions about his work from his portfolio and CV.

Early life

Mahmoud Sayed Youssef Kotb was born on 23 November 2000 in El Fath City, in the Assiut Governorate of Upper Egypt, and grew up in Egypt.

At school his strongest interest was physics. His physics teachers took to calling him أينشتاين مصر — "Egypt's Einstein" — and the nickname stayed with him; years later he gave it to his YouTube channel. What drew him to Einstein, by his own account, was less the historical figure than a way of thinking: curiosity, a readiness to question assumptions, and the habit of picturing a problem before formalising it. Physics shaped how he came to think about science and technology, and one line has stayed with him in particular — Einstein's remark, in a 1929 interview with the journalist George Sylvester Viereck, that "imagination is more important than knowledge".^[7]

Education

School

He attended the Lillian Trasher primary school (مدرسة لليان تراشر الابتدائية) from 2007 to 2013, then completed the preparatory (2013–2016) and secondary (2016–2019) stages at the Samih El-Said sports schools in Assiut (مدرسة سميح السعيد الرياضية الإعدادية and مدرسة سميح السعيد الثانوية الرياضية بأسيوط). He finished secondary school in 2019.^[a]

Assiut University

In September 2019 he entered the Faculty of Computers and Information at Assiut University, where he specialised in bioinformatics — a field he describes as applying data science and computational methods to biological problems. The programme combined computer science, statistics and biology, and taught him to treat a biological question as a computational one. He graduated in July 2023 with a B.Sc. in Computer and Information Science (Bioinformatics) and a grade point average of 3.53 out of 4.00.^[1][4] Projects from this strand of his work include a deep-learning classifier of cancer types from RNA-Seq gene-expression data and a tracker of SARS-CoV-2 genome mutations.^[4]

His interest in artificial intelligence deepened in the later years of the degree, during the COVID-19 pandemic — before ChatGPT made generative AI a public subject. Much of that study was self-directed: machine learning, pattern recognition and data mining, and beneath them the mathematics — linear algebra, probability, algorithms and discrete mathematics. What held his attention was less how to call a library than why a method works, and when it fails. Courses he completed include DeepLearning.AI's Machine Learning Specialization and Mathematics for Machine Learning: Linear Algebra on Coursera.^[4]

Student leadership

At university he became closely involved in student scientific life. In the 2021–22 academic year he served first as secretary of the scientific committee of his faculty's student union, and then as secretary of the Higher Scientific and Technological Committee of the Assiut University Student Union (أمين اللجنة العلمية العليا) — the university-wide committee drawn from the scientific secretaries of the individual faculties. He held the university-level post for about a year.^[4][8]

The role was one of coordination. He followed scientific activity across the faculties, acted as a link between students and the university administration, and organised or helped to organise more than seven competitions and events — among them contests in scientific innovation, science fiction, scientific writing and science video content, and seminars on astronomy and space, artificial intelligence, cybersecurity, scientific research and scientific writing. He worked closely with the university's astronomy club, where he was involved in the leadership, and helped to arrange educational visits connected with books, science and astronomy. The committee's work was recognised by outside institutions.^[4]

He regards the period as the one that taught him leadership, communication, event organisation, public speaking, and how to work with students from very different disciplines. He has led many of the collaborative projects he has worked on since, among them his graduation project and his master's research project.^[5][9] In the same years he was the student ambassador in Assiut for the Information Technology Institute (ITI), led a volunteer team at the USAID-supported University Center for Career Development, and attended Assiut University's Leaders Preparation Camp for 2021–22.^[4][10]

In his final year he was named the Ideal Student (الطالب المثالي) of the Faculty of Computers and Information for 2022–2023, taking first place in the faculty's Ideal Student competition — a recognition of academic results together with contribution to student life.^[4][10]

Teaching

Teaching is less a side activity in Alyosify's career than one half of a loop he describes this way: he learns, he builds, he teaches — and teaching exposes the gaps in his own understanding, which sends him back to study more deeply and to build better systems. It has also taught him what a degree rarely does: how to simplify without distorting, how differently learners think, and how to tell a technical story in public.

Einstein Misr

The channel began during the COVID-19 period, when he started explaining discrete mathematics to his classmates. The explanations grew into structured Arabic-language courses — discrete mathematics, algorithms and data structures, probability and statistics, and graph mining — published on YouTube as Einstein Misr and on Udemy. The channel's videos have been viewed more than 700,000 times, and the Udemy courses, among them Grokking Algorithms & Data Structures — in Arabic, Discrete Mathematics for Computer Science — in Arabic and Probability and Statistics — in Arabic, have enrolled more than 10,000 learners.^[2][3] In 2021–22 his team, White Hackers, won first place in the Science and Technology Content Competition at Assiut University.^[4] He intends to extend the channel into artificial intelligence and related technologies.

National Telecommunication Institute

He works as a machine-learning instructor at Egypt's National Telecommunication Institute (NTI). His first programme there, in the summer of 2025, was a 90-hour, laboratory-based course in supervised learning, principal component analysis and neural networks for engineering and computer-science students.^[1] He has since taught cohorts for the institute's branches in Al-Arish, Assiut and Minya, in a hybrid format; in all, he has delivered more than 300 hours of machine-learning instruction to more than 100 students. His courses run from foundations to advanced material, and he teaches the mathematics beneath each method — its intuition, the algorithm, its implementation, why it works and when it fails — rather than only how to call a library.

In June 2026 he presented his Automated Multimodal Agent, a system that turns PDF documents into narrated slide presentations, at Queen's University's AI in Teaching and Learning Forum. The university's special adviser to the provost on generative artificial intelligence wrote afterwards to thank him for the session.^[11][12]

Community teaching

His teaching has also taken the form of community service. In Assiut he volunteered with the Smile Makers Association for Orphan Sponsorship and Community Development (جمعية صناع البسمة لكفالة اليتيم), a charity affiliated with the Ministry of Social Solidarity, where he also served on the board. Through the association's summer educational club he gave workshops in mathematics, science, programming and introductory technology and AI to children from disadvantaged backgrounds, with the aim of making technical knowledge accessible to children who might not otherwise have access to it.^[4] He also taught programming and computational thinking to children for a short period with iSchool.

Military service

In the months between graduating and beginning his military service he completed a 200-hour machine-learning internship programme run by the Information Technology Industry Development Agency (ITIDA) under the Egypt Makes Electronics initiative (July–September 2023), and a diploma in full-stack development with .NET and Angular at Route Academy.^[1][4]

From February 2024 to March 2025 he completed Egypt's compulsory military service, working on systems at the National Company for Roads Building and Development, a company affiliated with the Egyptian Armed Forces that operates major roads in Egypt. Its toll-collection and vehicle-monitoring systems serve more than half of Egypt's vehicles across more than 70 per cent of the national road network.^[1] He supported the back-end systems behind these operations, co-developed an internal data-management and reporting application in C#, .NET, Entity Framework, SQL Server and Crystal Reports, and worked on a computer-vision system that reads vehicle licence plates to check whether a vehicle holds the subscription it needs to pass.

He counts the period among the most difficult of his life, and among the most formative. It taught him to work under pressure, to adapt to hard conditions, to take responsibility for systems that other people rely on, and to keep learning on his own when circumstances left little room for it.

AI and cybersecurity

From February to June 2025 he completed NTI's 420-hour AI in Cybersecurity programme: 288 hours of technical training in AI-driven threat detection, adversarial machine learning and intrusion-detection systems, and 132 hours of career development.^[4] It was followed by on-the-job training at e& Egypt (Etisalat) in June and July 2025, where he worked with security-operations tooling, memory forensics with Volatility, data-loss prevention and penetration-testing tools.^[1]

He came to the track with more experience in machine learning than in security, and chose it deliberately: he saw cybersecurity as a field where AI could make a distinctive contribution. The interest returned in his graduate projects on open-source intelligence.

Queen's University

In September 2025 he began an MSc in Artificial Intelligence at the School of Computing of Queen's University in Kingston, Ontario. He studied through Digilians, an Egyptian presidential initiative led by the Ministry of Communications and Information Technology that trains young Egyptians in advanced technology, including through professional master's programmes at international universities such as Queen's;^[13] his CV describes the award as the Digilians Presidential Scholarship.^[1] He completed the degree in 2026.

His coursework covered large language models, generative AI, deep learning, reinforcement learning, cloud computing and end-to-end machine-learning pipelines.^[1] For him the degree marked a change of mode — from an engineer who had largely taught himself the field to one doing formal graduate research. Dr. Rahatara Ferdousi, who taught him generative AI and supervised his master's project, singled out his "ability to connect theoretical ideas with practical implementation".^[14]

Research

MinimalLM

Language models often write more than a question needs, and every unnecessary token costs computation, time and energy. MinimalLM, his master's research project (CISC 898, 2026), asked whether a model can be made to say less at inference time — without retraining it or changing its architecture — while losing none of the content a reader needs.^[5]

Alyosify led the four-student team, which was supervised by Dr. Rahatara Ferdousi. The team built a training-free logits processor — a component that adjusts the model's next-token scores inside the decoding loop — combining a length term, an n-gram-repetition term and a semantic-redundancy term under an entropy gate. It was evaluated on three open-weight models (Mistral-7B-Instruct, Llama-3.1-8B-Instruct and Qwen2.5-3B-Instruct) across the MT-Bench, YapBench, XSum and SQuAD-v2 benchmarks, under a pre-registered protocol that treated "no loss of meaning" as a statistical non-inferiority hypothesis and also measured latency and GPU energy.^[5][1]

The project's most instructive result came from its choice of baseline. Compared with the simplest alternative — cutting the model's own answer to the same length — the processor's savings turned out to come mainly from ending answers earlier: in effect, it behaved as a stop controller. A survey of 23 families of training-free mechanisms, from logit penalties and activation steering to search and span selection, found none that chose what to keep better than matched-length truncation without access to oracle labels.^[5]

Besides leading the project, Alyosify worked across its tracks: the evaluation track — metrics, the language-model judge and claim auditing — the survey of mechanism families, the team's demonstration interface, and the consolidation and final audit of the evidence. As of September 2026 the project's results had not been published.^[5]

Research interests

He is currently researching the efficiency of language-model inference and token-efficient generation, AI agents and agentic systems, and the rigorous evaluation of generative models. Beyond them he is interested in world models, reinforcement learning for autonomous systems, AI and robotic systems, and quantum computing for machine learning.

Projects

Vitalism Solution

The subject of his graduation project was not new to him. In March 2022 his team won first place at the Smart Cities Hackathon, organised by Benha University with partners including Amazon Web Services and ElSewedy Digital, with a solution for monitoring the vital signs of patients in intensive care.^[15][4] His website records the result as first place among 77 teams from Egyptian public, private and national universities.^[16]

Vitalism Solution, his graduation project at Assiut University (2022–2023), estimates vital signs from ordinary video. It relies on remote photoplethysmography (rPPG): each heartbeat causes small, periodic changes in skin colour that a camera can record even when the eye cannot see them.^[9] Alyosify was the team leader of the six-student project, supervised by Dr. Ali Hussein Ahmed. He designed the video-based approach and worked on its core algorithms; the project site lists his roles as machine-learning specialist, signal processing, system integration and back-end logic.^[17]

The pipeline detects the face and skin regions with the Viola–Jones method, extracts a pulse signal from the video and filters it with wavelets. The team also compared deep-learning methods such as DeepPhys and PhysNet on the public UBFC-rPPG and PURE datasets, built desktop and Android applications, and compared the system's readings with those of conventional medical devices on visits to Assiut University's liver and heart hospitals.^[9][17] The project received an A+ grade, reached the semi-finals of Microsoft's Imagine Cup 2023, and was a finalist at the International Scientific and Engineering Innovation Competition (ISEIC) 2023, held at Egypt's Air Defense College, in which more than 800 teams took part.^[1][4]

Horus-OSINT and HORUS Sentinel

At Queen's, his interests in security and language models met in two related projects. Horus-OSINT (2026), built with two classmates for a graduate cloud-computing course, fine-tuned Meta-Llama-3-8B-Instruct with QLoRA on 159,826 records derived from the Global Terrorism Database and GDELT, and deployed it on Amazon Web Services as a conversational assistant for open-source intelligence, with data preparation in PySpark on Amazon EMR and the infrastructure defined in Terraform.^[18]

HORUS Sentinel (2026), developed with two teammates, makes that fine-tuned model the reasoning core of a multi-agent platform. Collection agents gather open-source intelligence, findings are correlated in a knowledge graph, and the self-hosted model — grounded through retrieval on the MITRE ATT&CK framework — drafts prioritised reports in which every claim can be traced to its source. The platform is passive and defensive by design: it uses only public data, requires a signed authorisation record, and leaves the final judgement to a human analyst.^[19]

AudioShield

AudioShield (2024) detects synthetic, "deepfake" speech. It converts a recording into a mel-spectrogram, classifies it as real or fake with a convolutional neural network, shows a Grad-CAM heatmap of what the model attended to, and produces a PDF report, all behind a Streamlit interface. He trained it on 81,000 spectrogram images, a dataset he published on Kaggle and Hugging Face.^[20]

Other work

His other projects include a self-supervised vision pipeline built on SimCLR, with a 42-experiment augmentation study, a semi-supervised SupCon extension and ONNX/FAISS retrieval, reaching 84.30% top-1 accuracy; a cloud autoscaler trained with the PPO and DQN reinforcement-learning algorithms in a custom Gymnasium environment; a reproducible repository-mining study of code-review effort on pull requests written by AI agents and by people; the Automated Multimodal Agent, which turns PDF documents into narrated slide presentations; and SCRAP, which predicts satellite collision risk from conjunction data available 48 hours or more before the closest approach.^[1][21]

Views on artificial intelligence

A recurring theme in Alyosify's work is using AI to extend human capability, and he values both halves of the job: understanding the science, and building systems people can use. He names mathematical understanding, novelty in research, careful experimentation, efficiency and practical impact among the things he cares about most.

His view of AI's future turns on imagination. As technology — and now AI itself — makes knowledge easier to reach, he argues, the distinctly human contribution shifts towards imagining what could exist. Human invention, in his account, has often been driven by necessity; he is interested in a future in which AI also helps people explore possibilities that are not yet necessities — to imagine, design, simulate and build technologies that improve life. He has described AI as potentially "the last invention of humanity", in the sense that ever more capable systems could accelerate much of what comes after them — an idea with a long history: in 1965 the statistician I. J. Good wrote that the first ultraintelligent machine would be "the last invention that man need ever make".^[22] Alyosify presents this as a personal perspective, not a prediction.

He has also thought about closer integration between people and AI — AI-assisted augmentation of memory and cognition, brain–computer interfaces, human–AI symbiosis — and about intelligent robotics, world models and autonomous scientific exploration, including AI and robotic systems able to explore environments such as other planets.

Long-term aims

The following are aspirations, not achievements. He wants to contribute to AI systems that expand what people can do and understand. One of his long-term ambitions is to help build a major Egyptian AI company that develops original technology and research and contributes internationally. He also intends to extend Einstein Misr into artificial intelligence and related technologies, and to continue research on efficient and agentic AI.

Recognition

• 2021–22First place, Science and Technology Content Competition, Assiut University (with the White Hackers team)
• 2022First place, Smart Cities Hackathon, Benha University
• 2022–23Ideal Student, Faculty of Computers and Information, Assiut University
• 2023Grade A+ for the graduation project, Vitalism Solution
• 2023Semi-finalist, Microsoft Imagine Cup (Vitalism Solution)
• 2023Finalist, International Scientific and Engineering Innovation Competition, ISEIC (Vitalism Solution)
• 2025Digilians scholarship for graduate study at Queen's University
• 2026Presenter, AI in Teaching and Learning Forum, Queen's University

He also holds more than 20 professional certifications, among them AWS Academy courses in machine learning and data engineering, Huawei's HCIP–AI and Microsoft Certified: Data Analyst Associate.^[1]

Chronology