Research Scientist - Data

Key Responsibilities

• Pioneer web-scale data collection and curation methodologies for LLMs and multi-modal foundation models. 
• Design and implement novel data synthesis pipelines for code, mathematics, and agentic reasoning datasets. 
• Trace the impact of data from pre-training to final model capabilities and create automated quality assessment frameworks for massive datasets 
• Design data recipes that maximize model capabilities across diverse domains. 
• Optimize data-model co-design for improved training dynamics. 
• Contribute to research papers and represent MBZUAI at industry conferences and events, showcasing the institution’s AI research and innovation. 

Academic Qualifications

• Minimum: Master’s in Computer Science, Data Science, or a related technical field, or equivalent practical experience required. 
• Preferred: PhD or equivalent research experience in Machine Learning, NLP, or Data Science with a focus on LLMs and data is preferred. 

Professional Experience

• Experience working with large language models, including evaluation, fine-tuning, and prompt engineering. 
• Strong Python development skills with a focus on research-grade code and scalable data pipelines. 
• Familiarity with collecting and processing large-scale datasets from open-source and web resources. 
• Demonstrated ability to work with ML infrastructure (e.g., model evaluation, optimization, debugging). 
• Proactive mindset with the ability to identify impactful research questions and execute on them with minimal supervision. 
• Effective communication and collaboration skills for working in cross-functional teams. 
Preferred   
• Prior research experience in areas such as web data curation and mixing, synthetizing complex datasets for training, LLM evaluation, post-training data, efficient inference, LLM-as-a-judge, tokenization. 
• Strong publication record in leading AI conferences (e.g., NeurIPS, ICLR, ICML, EMNLP) and/or prior contributions to open-source AI research or data tools. 
• Hands-on experience training language/mutli-modal models from scratch.