Founding AI Engineer

• Design model inputs that capture temporal context across video sequences rather than treating gameplay as a collection of static frames.
• Build and automate data pipelines that convert raw gameplay into high-quality, annotated training datasets.
• Optimise models for the latency, throughput and inference-cost requirements of real-time gameplay.
• Create evaluation suites and benchmarks that measure genuine agent capability.
• Take research ideas through experimentation, validation and production deployment.
• Help define the technical architecture, modelling strategy and engineering standards of an early-stage AI product.

• Substantial hands-on experience training or fine-tuning multimodal or Vision-Language Models.
• Strong understanding of transformer architectures and what happens beneath high-level model APIs.
• Experience with models or approaches such as LLaVA, CLIP, Flamingo or comparable multimodal architectures.
• Experience working with video data and modelling temporal context, potentially involving temporal attention, optical flow or video representation learning.
• A record of deploying deep-learning models into production environments where latency, reliability and cost matter.
• The ability to translate promising research into robust, maintainable systems.
• Comfort operating in a high-ownership, zero-to-one environment with incomplete information and rapidly changing priorities.
• A preference for shipping useful technology over producing research that remains theoretical.

An interest in games, simulation or complex interactive environments is important.

• Architect the core intelligence behind a new category of autonomous agent.
• Work on a difficult multimodal problem involving vision, video, reasoning and action.
• Own technical decisions across the full model lifecycle.

We’re partnering with an early-stage AI company building autonomous agents that understand, navigate and test complex digital environments.

The company began with a problem hiding in plain sight: modern video games have evolved into enormous, constantly changing worlds, while much of their quality assurance remains manual and repetitive. QA teams face increasingly complex releases, tight deadlines and thousands of possible player journeys - making comprehensive testing extraordinarily difficult.

Its technology uses vision-driven agents to interact with games as a player would, enabling human-like testing at machine scale. Rather than replacing experienced testers, the aim is to automate repetition and coverage so QA teams can concentrate on the creative, investigative and high-judgement work that produces exceptional games.

The business has already developed category-leading technology, secured backing from respected investors and AI ecosystem partners, and begun working with established game studios. Gaming is the initial proving ground, but the longer-term vision is considerably broader: extending the same agent technology into general software environments and, eventually, real-world robotics.

The opportunity

This is a founding-level position with ownership of the intelligence behind the company’s autonomous agents.

You’ll build the models that transform raw gameplay footage into an understanding of game state, intent and progression. That means owning the complete model lifecycle - from creating specialist datasets and designing temporal inputs to fine-tuning multimodal models, establishing evaluation frameworks and deploying them under real-time performance constraints.

This is suited to someone who combines research-level intuition with strong production engineering. You won’t be working on isolated experiments or publishing models that never leave a notebook. Your work will become the proprietary intelligence at the centre of a live product used to test real games.