Enterprise software companies typically spend €1–5 million annually on localisation, with 60–70% of these costs attributable to human post-editing of machine-translated content. The core challenge is not the quality of modern machine translation itself, but its inability to apply company-specific terminology, translation memories, style guides, and linguistic preferences that define an acceptable translation. As a result, human linguists must repeatedly correct otherwise accurate translations, creating a costly bottleneck that slows every software release.
This project will productise a multi-agent AI localisation platform based on the applicant's peer-reviewed MT Summit 2025 research prototype (TRL 4). The platform uses retrieval-augmented generation (RAG) to ingest an organisation's existing localisation assets and coordinates specialised AI agents for translation, terminology and style adaptation, automated post-editing, and quality assurance. Human translators remain in the workflow for lightweight validation, with the objective of reducing post-editing effort by at least 50% while maintaining equivalent translation quality.
Over 15 months, the project will focus on demonstrating commercial value rather than technical feasibility through structured industry pilots. Workday has confirmed its participation as an anchor customer (supported by a letter of support), with discussions also underway with Microsoft. The project will deliver validated customer demand, a pipeline of commercial opportunities, and the foundations for a DCU spin-out with early-stage investor readiness.
