From Eram Talent’s careers page.
This position is for building, deploying and evaluating Client's Arabic AI tutor for mathematics and computer science: preparing Arabic training data, fine-tuning, aligning and evaluating LLMs on Shaheen III and Ibex, and integrating the tutor into the learning platform for a pilot in at least 10 Saudi schools.
The position is part of the RDIA-funded project 5870, “An AI-Powered Arabic Tutoring Platform for STEM Education in the Kingdom.” Two engineers will be hired into this role.
Key Responsibilities
- Prepare Arabic training datasets from educational materials and consented platform interactions.
- Fine-tune, align, and evaluate commercial and open-source LLMs using 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 RLHF/DPO, educational technology, or mathematics/programming education is preferred.
Technical Skills
- Advanced Python and PyTorch.
- Hugging Face Transformers, PEFT, and TRL.
- Distributed GPU training using DeepSpeed or FSDP.
- LLM evaluation, reward modeling, and preference optimization.
- Arabic NLP, data pipelines, and OCR.
- Model serving and inference optimization using tools such as vLLM.
- APIs, Docker, Git, CI/CD, and production deployment.
