Research and learning

Foundation model pretraining

Foundation model pretraining is the work of designing broad model architectures and running the experiments that teach them general capability. AI companies need it because every later product, from fine-tuning to agents, inherits what pretraining gets right or wrong.

More companies want to train their own base models rather than rent someone else's, and the people who can do it well are few. Two of these roles are hiring now. Two are still emerging. The tricky part is evidence. Strong candidates show reproducible results and can explain why an experiment failed. Many have only fine-tuned existing models, which is a different skill. Training orchestration sits in distributed training infrastructure, so keep the briefs apart.

Hiring now 2Emerging 2

Roles in this family

4 roles

Where the people come from

Deep-learning research and reproducible model training

Not to be confused with. Training orchestration belongs to distributed training infrastructure.

Core skills

  • Experimental design
  • Deep learning
  • Transformer modelling
  • Data quality and provenance

Where it is used

  • Cross-sector AI platforms

Routes in and out

People move on to

More in Research and learning

← All new job families

Start a search

Hiring in foundation model pretraining?

Tell us the work you need done. We find the people already doing it, whatever their job title says.