I'm a Computer Science graduate with a strong passion for machine learning, multimodal AI, and autonomous systems. I enjoy transforming complex research ideas into real-world systems and building intelligent models that reason, perceive, and act.

Throughout my academic journey, I worked on multiple end-to-end AI projects, including Vision-Language-Action models for autonomous driving, OCR systems, and speech synthesis models. I’m driven by research, engineering, and pushing AI from theory to production.

A research-driven Vision-Language-Action (VLA) framework designed to enable multimodal large language models to perceive driving scenes, reason about them, and generate vehicle control commands.

• Designed a unified architecture combining vision, language reasoning, and ego-vehicle state
• Fine-tuned Qwen2.5-VL for autonomous driving tasks
• Generated structured outputs: trajectory, speed, and steering predictions
• Enabled explainable driving decisions using language grounding
• Trained and evaluated on DriveGPT4, LingoQA, DriveLM datasets

Tech: PyTorch, Qwen2.5-VL, Multimodal LLMs, Autonomous Driving, Computer Vision

A published research project focused on improving Arabic OCR accuracy through large-scale synthetic data generation and model fine-tuning.

• Built a dataset of ~1 million Arabic sentences with and without tashkeel
• Generated synthetic OCR images using TRDG with multiple Arabic fonts
• Fine-tuned Tesseract 5 Arabic models (ara.traineddata)
• Evaluated performance using CER and WER benchmarks
• Achieved significant WER improvements on clean datasets

📎 Paper: https://ieeexplore.ieee.org/document/10928060

Tech: Tesseract OCR, Arabic NLP, Dataset Engineering, OCR Evaluation

A beta Text-to-Speech model built on the XTTS v2 architecture.

• Focused on voice quality and performance
• Published and documented the model for public use

Tech: Speech Processing, Deep Learning, Model Training

A Python package for generating OCR training datasets from text.

• Supports multilingual text rendering with custom fonts
• Includes noise augmentation for OCR robustness
• Published on PyPI with full documentation (MkDocs)

Tech: Python Packaging, Data Augmentation, OCR Pipelines

• Fine-Tuning an Arabic OCR Model using Tesseract 5
  IEEE Xplore
  https://ieeexplore.ieee.org/document/10928060

• Computer Vision & Image Processing
• OCR Systems & Document AI
• Vision-Language Models (VLMs, MLLMs)
• Vision-Language-Action (VLA) Models
• Multimodal Dataset Curation
• Model Fine-Tuning & Evaluation

• PyTorch
• LLaMAFactory
• MkDocs
• Git & GitHub
• Linux

Autonomous Driving • Multimodal AI • Large Language Models • Applied Machine Learning • AI Research

• GitHub: https://github.com/OmarSamirz
• LinkedIn: https://www.linkedin.com/in/omar-samir-8415b2285/
• Email: omarsamir1300@gmail.com