AI Research Engineer: Data Pipelines & Agent Training

Dex; Greater London; Hybrid; GBP 90,000 - 150,000; Full time

Job summary

Dex is partnering with Frontier labs to recruit a Member of Technical Staff in London or Paris. You will own ambiguous research-engineering problems across data curation, model fine-tuning, and agent execution harnesses, applying SFT and RL and turning prototypes into hardened systems.

You will design training-data pipelines, fine-tune/post-train models, develop agent harnesses, and collaborate with software and curriculum teams to translate open-ended questions into production-ready solutions.

Qualifications

• Hands-on experience fine-tuning or post-training models using SFT or RL.
• Strong grounding in AI/ML fundamentals beyond orchestration.
• Proficiency in Python and building production-grade code.
• Ability to translate ambiguous research questions into technical plans.
• Based in or willing to work hybrid from London or Paris.

Responsibilities

• Design, curate, and scale training-data and synthetic-data pipelines for agent training.
• Fine-tune and post-train models using supervised fine-tuning and RL methods.
• Develop and harden agent harnesses and sandboxed execution environments.
• Translate open-ended AI questions into technical plans and production-ready systems.
• Drive technical decisions and coordinate complex engineering efforts across the team.

Skills

Fine-tuning / post-training models

Python programming

AI/ML fundamentals

System design & evaluation

Hybrid work readiness London/Paris

Tools

PyTorch

TensorFlow

RLHF

Job description

Dex is partnering with Frontier labs to recruit a Member of Technical Staff in London or Paris. You will own ambiguous research-engineering problems across data curation, model fine-tuning, and agent execution harnesses, applying SFT and RL and turning prototypes into hardened systems.

You will design training-data pipelines, fine-tune/post-train models, develop agent harnesses, and collaborate with software and curriculum teams to translate open-ended questions into production-ready solutions.