About the role
Training frontier AI agents requires interactive digital environments where models can act, fail, and learn across real software. In roughly a year, this seed-stage research lab has scaled to over 50 engineers across London and Paris, partnered with multiple frontier AI labs, and secured backing from top-tier investors and founders. The core constraint on agent performance is data: acquiring, sanitising, and reliably serving massive, messy real-world datasets to the engineers building those training environments.
Responsibilities
•Architect and operate scalable ingestion, processing, and storage systems that handle messy, high-volume datasets from real-world software environments.
•Build automated validation, sanitisation, and PII-redaction pipelines that make raw data immediately usable for AI training environments.
•Make pragmatic technical trade-offs across storage, retrieval latency, observability, system reliability, and cloud costs.
•Establish reusable data platform patterns that other software and AI engineers can build on, eliminating the need to solve dataset ingest on an ad-hoc basis.
•Debug unfamiliar production failures across the stack and deploy modern tooling to keep engineering feedback loops tight.
Requirements
•Proven track record designing, shipping, and operating substantial production data systems end to end, with personal ownership over the architectural decisions and outcomes.
•Strong software engineering fundamentals with the ability to dive deep into unfamiliar codebases and debug complex, distributed failure modes.
•Direct experience working through large, messy, or adversarial datasets where ingestion, validation, or sanitisation posed genuine technical bottlenecks.
•Demonstrated ability to take ambiguous requirements, define the necessary system architecture, and ship production systems without detailed functional specs.
•Based in or able to work hybrid from London or Paris.
Conditions
Not specified in the posting.
