About the Role
This position involves building, deploying, and evaluating an Arabic AI tutor for mathematics and computer science. Responsibilities include preparing Arabic training data, fine-tuning, aligning and evaluating large language models (LLMs) on Shaheen III and Ibex, and integrating the tutor into a learning platform for a pilot program in at least 10 Saudi schools. This role is part of the RDIA-funded project 5870, "An AI-Powered Arabic Tutoring Platform for STEM Education in the Kingdom," with two engineers being hired for this purpose.
Ready to apply for roles like this?
Unlock the company name and direct application link. Subscribers get instant access to fresh jobs across Dubai, Abu Dhabi and Riyadh, many with visa support.
Unlock employer & apply directly
Key Responsibilities
- Prepare Arabic training datasets from educational materials and consented platform interactions.
- Fine-tune, align, and evaluate commercial and open-source LLMs using the Shaheen III and Ibex computing platforms.
- Develop Arabic STEM benchmarks, preference-training methods, and student-safety guardrails.
- Integrate AI tutoring capabilities into the learning platform, including Arabic right-to-left interfaces.
- Deploy and optimize models for reliable, low-latency real-time interactions.
- Support school pilot implementation, evaluation, and teacher feedback collection.
- Ensure student data privacy and compliance with Saudi data protection requirements.
- Contribute to research publications, technical reports, and project deliverables.
Qualifications & Experience
- MSc in Computer Science, Artificial Intelligence, or a related field; alternatively, a BSc with at least one year of relevant experience.
- Hands-on experience in LLM fine-tuning, evaluation, and AI application deployment.
- Professional proficiency in Arabic and English.
- Experience with Reinforcement Learning with Human Feedback (RLHF)/Direct Preference Optimization (DPO), educational technology, or mathematics/programming education is preferred.
Technical Skills
- Advanced proficiency in Python and PyTorch.
- Familiarity with Hugging Face Transformers, Parameter-Efficient Fine-Tuning (PEFT), and Teaching Reinforcement Learning (TRL).
- Experience in distributed GPU training using DeepSpeed or Fully Sharded Data Parallel (FSDP).
- Knowledge of LLM evaluation, reward modeling, and preference optimization.
- Skills in Arabic Natural Language Processing (NLP), data pipelines, and Optical Character Recognition (OCR).
- Proficiency in model serving and inference optimization using tools such as vLLM.
- Familiarity with APIs, Docker, Git, Continuous Integration/Continuous Deployment (CI/CD), and production deployment.
