The Role
This role exists for one reason: to build AI systems that get better on their own.
We're looking for an AI Tooling Specialist with a deep focus on recursive self-learning — the design and implementation of systems that evaluate their own outputs, identify failure modes, and update their own behaviour without requiring constant human intervention. You'll own this capability end-to-end at Team Wheel, from architecture through to production.
If you've spent time thinking seriously about how to make AI systems that genuinely improve over time — not just ones that are prompted better — this is the role for you.
We believe in paying fairly and being upfront about it. For this role, we're offering a base salary of £80,000–£110,000 depending on experience, with additional performance-related pay. We know the people we're looking for have options, and we want them invested in what we're building together so we are open to building a creative package for the right individual, possibly including shadow equity.
What You'll Do
Design and Build Self-Improving Systems
-
Architect recursive learning loops in which AI outputs are automatically scored, critiqued and fed back into subsequent generations
-
Build self-critique and self-refinement pipelines where models evaluate and iteratively improve their own responses before delivery
-
Implement Constitutional AI-style or critic-model approaches to drive automated quality improvement without human-in-the-loop on every iteration
-
Design memory and state mechanisms that allow systems to accumulate and apply learnings across sessions and tasks
Build the Evaluation Layer
-
Develop robust, automated evaluation frameworks (evals) that measure output quality across multiple dimensions
-
Create feedback signal pipelines — combining automated scoring, preference signals and sparse human feedback — to drive continuous improvement
-
Build regression detection so the system flags when it is getting worse, not just when it is getting better
-
Instrument all systems for observability: you'll know at any point whether the loop is working
Automate the Optimisation Process
-
Implement automated prompt and chain optimisation (DSPy, TextGrad, or similar) so that the system tunes itself rather than relying on manual prompt engineering
-
Build processes that generate, test and promote improved instructions or configurations autonomously
-
Work towards systems that can propose and validate their own architectural changes within defined constraints
Integrate and Deploy
-
Build integrations with the platforms Team Wheel works with (Workday and other HR SaaS)
-
Own reliability and performance in production — these are live systems, not experiments
-
Document the architecture, the learning mechanisms, and the failure modes clearly
What We're Looking For
Essential
-
Strong Python and hands-on experience building with LLMs at a production level
-
Demonstrable experience designing recursive or iterative self-improvement loops in AI systems
-
Deep familiarity with evaluation design — automated scoring, preference modelling, critic models
-
Experience with prompt/chain optimisation frameworks (DSPy, TextGrad, or equivalent)
-
Solid understanding of agent architectures and multi-step reasoning pipelines
-
You can articulate clearly why a self-learning loop failed — not just that it did
Desirable
-
Background in or strong understanding of RLHF, DPO, or related preference optimisation techniques
-
Experience with Constitutional AI, self-critique, or debate-style architectures
-
Exposure to enterprise HR platforms or complex API integrations
-
Published work, open source contributions, or personal projects in self-improving AI systems
Why Join Team Wheel
-
Own a genuinely novel capability inside a growing consultancy
-
Production systems from day one — not research, not prototypes
-
Work alongside domain experts who understand the HR tech context deeply
-
A culture that stays close to the frontier and invests in doing this properly
-
Competitive salary and flexible working
Team Wheel is an equal opportunities employer. We welcome applications from candidates of all backgrounds.
#J-18808-Ljbffr
