SAMAALI is an AI-powered application designed to support students in memorizing and reviewing
their lessons without the need for a human to recite and assess their memorization. The system
begins by indexing textbooks and applying optical character recognition (OCR) technology to
extract the texts to be recited. The system then uses Whisper for automatic speech recognition
(ASR) and transcribes the text to be recited.
The platform evaluates students’ verbal responses to enhance learning through speaking. It also
integrates AraT5 to automatically generate questions based on the extracted key concepts,
providing a dynamic and personalized learning experience.
The backend is built using Django and integrates Firebase for instant notifications and Twilio for
secure SMS and email verification. A dedicated dashboard enables teachers and administrators to
manage user progress, review generated questions, and monitor OCR and ASR data with complete
clarity.
The mobile app allows students to review lessons, access generated questions, listen to audio
content, and track their performance—creating an integrated and smart learning environment.

I developed a robust OCR and NLP pipeline to extract Arabic text from scanned images and PDF
documents. Instead of relying solely on traditional OCR engines, I integrated advanced vision
language models—Idefics2 for visual-text recognition and Mistral for linguistic post-processing and
interpretation. The pipeline included preprocessing steps such as noise reduction, thresholding,
and line detection to enhance input quality. Extracted text was refined using a correction layer that
leveraged n-gram similarity and contextual rules to address common OCR errors. This system was
particularly effective for digitizing historical and academic Arabic documents, supporting reliable
text retrieval, archiving, and downstream NLP tasks.

I developed an advanced system for automatically generating educational questions from Arabic
content using the AraT5 Transformer model. The process began by extracting key phrases from
Arabic textual materials using ARAELECTRA in combination with syntactic dependency parsing,
enhancing the model’s understanding of contextual and grammatical relationships. To ensure
clean and structured input, I applied a comprehensive preprocessing pipeline—including
tokenization, diacritic normalization, and text cleaning. Following this, I fine-tuned the AraT5
model on a custom Arabic QA dataset, enabling it to generate coherent, contextually relevant,
and grammatically accurate questions aligned with the extracted core concepts.
In a subsequent stage, I incorporated Large Language Models (LLMs) to generate multiple
choice questions (MCQs) with corresponding answers, ensuring that distractor options were
contextually plausible and carefully designed to avoid biasing students toward the correct
answer. This integration provided balanced, fair, and pedagogically sound assessments tailored

Developed an AI-powered Chabot using a fine-tuned Llama language model integrated with
LangChain to provide real-time assistance in a tourism management system. LangChain was
leveraged to manage prompt engineering, handle conversation memory, and enable retrieval
augmented generation (RAG) for accurate and context-aware responses. The Chabot supported
multilingual interactions (Arabic and English) and addressed frequently asked questions, travel
guidance, and support services. Intent recognition and rule-based dialogue flows were
combined with LangChain’s chaining and agent capabilities to maintain structured control, while
fallback handling ensured graceful responses for unrecognized queries.

A comprehensive car listing and selling platform with a simple, user-friendly interface in Arabic
and English, with support for additional languages. It ensures a smooth browsing experience
with non-intrusive ads.
Users can explore dealership listings, compare prices, and access essential details to find the
right vehicle. Direct support is available via tickets or WhatsApp, with automated notifications
and emails for verification and updates.
Secure online payments streamline booking and purchasing. For dealerships, an advanced
control panel enables vehicle management, promotions, and performance tracking through
analytics for better decision-making

This project is designed to efficiently manage class reservations, instructor schedules, exams,
placement tests, and courses in an integrated manner.
It ensures that schedules and engagements within the institute do not overlap, allowing for a
seamless and conflict-free administrative process.
The system provides essential tools such as filtering options, notifications, and reminders, which
are sent through the control panel to keep users informed of important deadlines and
appointments. By streamlining administrative operations, the system enhances efficiency,
accuracy, and overall management within the institute.
Additionally, a dedicated mobile application allows students to track lessons, listen to audio
recordings, view test results, and access various course-related services.