AI Engineer – LLM Data

Institute of Foundation Models

Role

Key responsibilities

• Build and improve large-scale datasets for LLM training and support time-sensitive data and model development tasks.
• Develop extraction, cleaning, filtering, deduplication, quality-scoring, transformation, and dataset-composition pipelines.
• Design LLM data-synthesis and augmentation workflows for generation, filtering, refinement, and quality control.
• Research improvements in training-data quality, diversity, coverage, and efficiency; run experiments linking training data to model performance.
• Develop scalable data-processing tools; analyze datasets for quality issues, bias, duplication, distribution gaps, and improvement opportunities.
• Collaborate with research and engineering teams and document datasets, experiments, methods, and findings for reproducibility.

Required experience

• Strong Python programming and experience building reliable engineering or research workflows.
• Hands-on experience with machine learning, deep learning, NLP, or LLMs; understanding of how training data is used in pre-training and/or post-training.
• Experience processing and analyzing large-scale datasets, including filtering, transformation, and quality control.
• Ability to investigate quality issues, design experiments, and make data-driven technical decisions.
• Familiarity with ML/LLM tools such as PyTorch, Hugging Face, vLLM, or similar.
• Strong problem-solving, communication, and collaboration skills in a fast-paced research and engineering environment.

Preferred experience

• LLM pre-training, continued/mid-training, supervised fine-tuning, preference optimization, reinforcement learning, or other post-training data workflows.
• Synthetic data generation and quality evaluation; LLM evaluations, benchmarks, model training, or fine-tuning experiments.
• Understanding of data quality, mixture, diversity, and scaling effects on foundation models.
• Distributed data processing or compute infrastructure; research projects, open-source contributions, publications, or hands-on modern foundation model work.