For AI developers
Operational data from European industry, not from the open web.
Oreline sources proprietary datasets from established companies across twelve sectors, from construction and engineering to film, publishing, software and field services. Every owner we introduce has been qualified, every dataset has been scoped for rights and personal data, and every licence is concluded directly with you.
What we source
Data created in the course of real commercial work, often with matched records across a single project: the drawing, the quote, the machining file and the photograph of the result.
- CAD & BIM
- DWG, DXF, STEP, IGES, Revit, IFC and native 3D formats, with revision history
- Commercial documents
- Quotes, take-offs, bills of quantities, specifications, proposals (PDF, XLSX, DOCX)
- Manufacturing
- CNC programs, cut lists, nesting files, process parameters, work instructions
- Imagery
- Site, installation, inspection and defect photography (JPEG, RAW), drone surveys
- Video
- Raw rushes, finished originals, dashcam and operational footage (ProRes, MXF, H.264)
- Audio & speech
- Podcasts, radio archives, voice recordings, music stems, with transcripts
- Expert text
- Reports, opinions, methodologies, technical documentation, largely in French
- Conversations
- Support tickets, chat transcripts and knowledge bases, fully anonymised
- Code
- Proprietary repositories with commit history, tickets and code review threads
- Time series & logs
- Machine sensor data, maintenance logs, delivery and incident records
How we qualify supply
Introductions are only made once these checks are complete. You receive fewer, better-prepared counterparties.
- Counterparty check
- Each data owner is an established, registered company. We verify its identity, trading history and the decision-maker we are dealing with.
- Chain of title review
- We review with the owner where the material comes from and flag third-party content (external architects' drawings, licensed music, co-produced footage) before any offer.
- Personal data scoping
- We identify where personal data sits (title blocks, layers, metadata, faces, voices) so that anonymisation requirements can be written into the licence.
- Volume and consistency
- We document formats, years covered, approximate volumes and how files relate to one another, so your team can scope a sample request precisely.
How an engagement works
Oreline does not hold, resell or sublicense data. We prepare the relationship; the licence is yours and the owner's.
Share your requirements
Domains, formats, volumes, languages and licence terms you are working to.
Receive qualified introductions
A short profile of each matching data owner, with the decision-maker ready to engage.
Evaluate and contract directly
Your team reviews samples under NDA and contracts directly with the data owner.
Deliver
Extraction and delivery follow your standard process and the terms of the licence.
Why European, why now
European SMEs hold decades of trade-specific records, produced to strict regulatory and quality standards, and largely absent from public training corpora. Much of it is in French and other European languages.
They are also cautious. A licence that respects the GDPR, the EU AI Act's transparency obligations on training data and the owner's commercial interests is the only way this supply comes to market. That is the standard we work to.
Tell us what your models need
Write to sourcing@orelinedata.com with the domains, formats and volumes you are targeting. We respond within two business days.