AI Product Engineer — End-to-End AI Certification Platform

Dex; Greater London; On-site; GBP 75,000 - 100,000; Full time

Job summary

Dex partner company is building the first AI-native certification body, assessing AI systems for security, reliability and intended behavior. You will join a founding team as a Product Engineer, owning features from UI to AI-agent infrastructure, shaping the processes for certification.

You will design and implement full-stack features, build the core agent infrastructure and evaluation frameworks, and define safeguards like sandboxing and hallucination containment to impact audit outcomes and

Qualifications

• Shipped real systems with substantive decisions on agent harnesses, evals, hallucination containment, sandboxing or multi-component system/tool design.
• Proven ability to ship across interface, backend, data, and deployment layers.
• Reasons from first principles, choosing tools deliberately rather than following a playbook.

Responsibilities

• Design and implement full-stack product features, from user interface to backend services.
• Build the core AI agent infrastructure and evaluation frameworks for system certification.
• Define and implement critical safeguards for AI systems, including agent sandboxing and hallucination containment.
• Drive product development from initial concept through to production deployment, ensuring direct impact on audit outcomes.
• Contribute to the technical architecture and foundational engineering practices of a nascent team.

Skills

Agent sandboxing

Frontend & Backend

Evaluation frameworks

Open-source evidence

Job description

Dex partner company is building the first AI-native certification body, assessing AI systems for security, reliability and intended behavior. You will join a founding team as a Product Engineer, owning features from UI to AI-agent infrastructure, shaping the processes for certification.

You will design and implement full-stack features, build the core agent infrastructure and evaluation frameworks, and define safeguards like sandboxing and hallucination containment to impact audit outcomes and