Research Engineer Roles - AI Training Data

Data Scorecard is hiring research engineers to build the data infrastructure layer for AI training. Both openings focus on making data curation measurable across pre-training and post-training, then turning successful methods into product features.

Open roles

• Founding Research Engineer - AI Training Data: a full-time founding role owning data curation as a measured discipline across every stage of training.
• Research Engineer Intern - AI Training Data: a full-time or part-time internship doing data curation research alongside the founders.

Founding Research Engineer

• Build metrics for dataset quality, provenance, safety, and coverage that help predict model behaviour.
• Diagnose data issues such as contamination, uneven multilingual coverage, long-tail gaps, and difficulty mismatches.
• Design and test curation interventions including pruning, filtering, synthetic augmentation, and relabelling.
• Turn recent research into practical training and evaluation loops that validate data hypotheses.
• Ship useful methods as product features and share findings through technical reports or papers.

What they are looking for

• Strong machine learning and deep-learning fundamentals.
• Enough software engineering and PyTorch or Jax experience to run ML experiments and build production prototypes.
• Hands-on experience training or evaluating LLMs or vision-language models.
• Experience with data curation, pruning and selection, synthetic data, curriculum learning, dataset distillation, or large-scale language or multimodal training.
• Comfort reading ML research, identifying promising ideas, and implementing them.
• Ability to drive applied research independently in a fast-moving early-stage environment.

Nice to have

• Post-training experience such as SFT, preference optimisation, RLVR, or reward modelling.
• Multilingual or multimodal data work.
• Multi-GPU or distributed training experience.
• Open-source or Hugging Face contributions.
• Public technical writing or published research.

Research Engineer Intern

• Measure dataset quality, provenance, safety, and coverage metrics that predict downstream model performance.
• Diagnose where data will hurt a model, including decontamination, difficulty annotation, multilingual asymmetries, and long-tail gaps.
• Run curation interventions such as filtering, deduplication, and synthetic augmentation, then build eval harnesses to measure their effect.
• Read recent research, reproduce promising methods, and push them beyond the original paper.
• Help turn working methods into product features and share findings as a technical report or paper.

What they are looking for

• Solid Python and working knowledge of the ML stack, including PyTorch or Jax and Hugging Face.
• Exposure to LLMs or vision-language models through coursework, research, projects, fine-tuning, evaluation, or data pipelines.
• Comfort reading an ML paper and turning it into a working experiment.
• Evidence you can build and reason about experiments, such as a repo, paper, project, or competition.
• Current students, recent grads, and strong self-taught engineers are welcome.

Nice to have

• Post-training experience such as SFT, preference optimisation, RLVR, or reward modelling.
• Multilingual or multimodal data work.
• Synthetic data experience.
• A shipped side project, tool, or demo people used.
• Open-source or Hugging Face contributions.
• Research experience, publications, or technical writing.