Summary

Train and improve Cosine's Lumen Enterprise models for software engineering, working across pre-training, fine-tuning, reinforcement learning, and distributed training. This London-based role is fully in-office five days a week.

Responsibilities

• Run pre-training, fine-tuning, and reinforcement-learning experiments
• Optimize training methods for frontier models
• Diagnose distributed-training bottlenecks and communication overhead
• Train models on repository-level coding tasks
• Develop reinforcement-learning loops using engineering feedback
• Investigate planning failures, brittle fixes, and reward hacking