Senior AI Engineer — Build Real-World Agent Training

Dex; Greater London; On-site; GBP 90,000 - 130,000; Full time

Job summary

Dex is partnering to build the digital environments where AI agents learn by doing, actively navigating software and improving through trial and error. The team generates experiential data that trains the next generation of AI in collaboration with frontier AI labs.

The role involves designing data generation methods, fine-tuning models with SFT/RL/RLHF, and building robust evaluation loops to measure and improve performance in interactive environments.

Qualifications

• Directly fine-tuned or post-trained models, with ability to articulate methods, ownership, and evaluation.
• Strong command of AI/ML fundamentals, including SFT, RL, RLHF, or similar post-training methods.
• Proficient in production-grade Python, writing clean, testable code for complex engineering challenges.
• Ability to tackle highly ambiguous problems from first principles without a predefined playbook.
• Proven track record of owning significant technical outcomes across research, data, and engineering.

Responsibilities

• Design and implement methods for generating and curating high-quality experiential data for AI agent training.
• Fine-tune and post-train large models using supervised fine-tuning, reinforcement learning, and automated optimisation techniques.
• Build robust evaluation and feedback loops to measure and improve model and agent performance in interactive environments.
• Develop agent harnesses and sandboxed environments that connect AI models to complex software tasks.
• Take ambiguous research questions from initial concept through to production-grade, observable AI systems.

Skills

Model fine-tuning/post-training

SFT, RL, RLHF

Production-grade Python

Ambiguity resolution

Ownership of technical outcomes

Tools

Python

Job description

Dex is partnering to build the digital environments where AI agents learn by doing, actively navigating software and improving through trial and error. The team generates experiential data that trains the next generation of AI in collaboration with frontier AI labs.

The role involves designing data generation methods, fine-tuning models with SFT/RL/RLHF, and building robust evaluation loops to measure and improve performance in interactive environments.