AI/LLM Engineer

About the Role

This position involves designing, deploying, and assessing an Arabic-language AI tutor targeting mathematics and computer science education. Responsibilities include preparing Arabic training datasets derived from educational content and user interactions, fine-tuning and aligning large language models (LLMs) on specialized computing platforms, and integrating the tutoring system into an educational platform for pilot testing in at least 10 Saudi Arabian schools. The role is part of the RDIA-funded project 5870 titled "An AI-Powered Arabic Tutoring Platform for STEM Education in the Kingdom." Two engineers will be recruited for these responsibilities.

Key Duties

• Create Arabic training datasets from curricula and authorized platform data.
• Conduct fine-tuning, alignment, and evaluation of commercial and open-source LLMs on Shaheen III and Ibex computing resources.
• Develop benchmarks in Arabic STEM education, methods for preference training, and safeguards to ensure student safety.
• Embed AI tutoring features into the learning system, supporting Arabic right-to-left user interfaces.
• Deploy and maintain optimized models ensuring dependable, low-latency, real-time interactions.
• Assist with pilot implementations in schools, including evaluations and collecting teacher feedback.
• Ensure compliance with privacy and Saudi data protection laws to protect student information.
• Contribute to academic publications, technical documentation, and fulfill project deliverables.

Candidate Profile and Qualifications

• Master's degree in Computer Science, Artificial Intelligence, or relevant field; candidates with a Bachelor’s degree plus at least one year of related experience may also apply.
• Practical experience in LLM fine-tuning, assessing, and deploying AI applications.
• Fluency in both Arabic and English at a professional level.
• Experience or familiarity with Reinforcement Learning from Human Feedback (RLHF) or Direct Preference Optimization (DPO), educational technology, or teaching mathematics/programming is an advantage.

Technical Expertise Required

• Expertise in Python and PyTorch frameworks.
• Proficiency with Hugging Face Transformers, Parameter-Efficient Fine-Tuning (PEFT), and TRL libraries.
• Experience in distributed GPU training techniques such as DeepSpeed or Fully Sharded Data Parallel (FSDP).
• Skills in evaluating LLMs, reward modeling, and fine-tuning preference optimization.
• Familiarity with Arabic Natural Language Processing (NLP), data processing pipelines, and Optical Character Recognition (OCR).
• Know-how in model serving and inference improvements, for example using vLLM.
• Competence with APIs, containerization tools like Docker, version control with Git, along with continuous integration/deployment (CI/CD) practices for production environments.