Computer Vision Engineer

Key Responsibilities

• Fine-tune pre-trained OCR models (e.g., DeepSeek-OCR, Dots-OCR, EasyOCR) on Arabic handwritten datasets.
• Collect, annotate, and preprocess Arabic handwritten data including text, forms, and documents.
• Design and implement data augmentation pipelines tailored for Arabic script variability (diacritics, connected letters, cursive styles).
• Evaluate model performance using CER (Character Error Rate) and WER (Word Error Rate) metrics.
• Integrate trained OCR models into backend APIs or document processing pipelines.
• Continuously improve model accuracy through iterative training, hyperparameter tuning, and error analysis.
• Develop real-time hand gestures and sign language recognition models using video and image data.
• Implement pose estimation techniques (e.g., MediaPipe, OpenPose) to extract hand and body keypoints.
• Build and train classification models (CNNs, LSTMs, Transformers) for static and dynamic sign recognition.
• Work with domain experts and the deaf/hard-of-hearing community to gather and validate sign language datasets.
• Optimize models for real-time performance on edge devices or web/mobile applications.
• Document model architecture, training procedures, and deployment steps for the engineering team.
• General Engineering Responsibilities
• Collaborate with backend engineers to deploy models using Fast API, Flask, or similar frameworks.
• Write clean, well-documented, and testable Python code following best engineering practices.
• Participate in code reviews, sprint planning, and technical architecture discussions.
• Stay up to date with the latest research in OCR, NLP for Arabic text, and gesture recognition.
• Contribute to internal tooling for data labeling, model monitoring, and performance dashboards.

Required Qualifications

• Bachelor's or master’s degree in computer science, Electrical Engineering, AI, or a related field.
• 2–3 years of hands-on experience in computer vision or machine learning engineering.
• Strong proficiency in Python; experience with PyTorch or TensorFlow for model development.
• Solid understanding of deep learning architectures: CNNs, RNNs/LSTMs, Transformers, and Attention mechanisms, QLoRa.
• Experience with image preprocessing, augmentation libraries (Augmentations, OpenCV, PIL).
• Familiarity with OCR pipelines and text recognition challenges.
• Experience with fine-tuning models on custom datasets and transfer learning techniques.
• Ability to work with version control (Git) and experiment tracking tools (MLflow, Weights & Biases).
• Preferred Qualifications
• Prior experience working with Arabic NLP or Arabic OCR datasets is a strong plus.
• Experience with sign language datasets or human pose/gesture estimation frameworks (MediaPipe, OpenPose).
• Familiarity with Hugging Face Transformers and pre-trained vision-language models (DeepSeek-OCR, Dots-Ocr, EasyOCR).
• Knowledge of model quantization, pruning, and ONNX/TensorRT deployment for inference optimization.
• Experience building REST APIs for model serving (FastAPI, Flask).
• Exposure to annotation tools such as Label Studio, CVAT, or Roboflow.
• Published research, open-source contributions, or Kaggle competition experience.