Industries
Wherever real work leaves a record, there is data AI developers need.
AI developers are moving beyond public web data towards the records that show how professional work is actually done. These are the twelve sectors where we see sustained buyer demand today.
- Construction & fit-outAutoCAD DWG / DXF shop drawings and their revisions, Quantity take-offs, quotes and bills of quantities
- Engineering & CAD3D models and 2D production drawings, Bills of materials and calculation notes
- Film & documentaryRaw footage and unedited rushes, Original productions with clear chain of title
- Architecture & real estateFloor plans, BIM models (Revit, IFC) and building permit files, Renderings and before / after project photography
- Manufacturing & qualityVisual inspection images with defect labels, Process parameters and machine sensor time series
- Field services & maintenanceIntervention reports and diagnostic notes, Before / after photographs of repairs
- Legal, accounting & consultingTemplates, model clauses and standard documents, Anonymised opinions, memos and reports
- Publishing, photography & audioBook, magazine and press archives, Photo libraries with captions and metadata
- Software & ITProprietary source code with its commit history, Tickets, bug reports and their resolutions
- Customer support & salesSupport tickets and chat transcripts, Knowledge bases and response macros
- Agriculture & environmentField, crop and plant disease imagery, Yield, soil and treatment records
- Logistics & transportDelivery, route and incident logs, Dashcam and warehouse camera footage
Construction & fit-out
Joinery, stone and worktop fabrication, interior fit-out, finishing trades, general contractors.
Shop drawings paired with the quote, the machining file and the photo of the finished job show how a design becomes a built object. That end-to-end trace is almost never available on the open web.
Typical datasets
- AutoCAD DWG / DXF shop drawings and their revisions
- Quantity take-offs, quotes and bills of quantities
- CNC and machining files
- Site and installation photography
What we check first
- Title blocks, layers and file properties often contain client names and site addresses: these must be removed.
- Drawings supplied by external architects or engineers may not be yours to license.
- Photographs may show clients' homes and people: faces and identifying details are blurred or excluded.
Engineering & CAD
Design offices, mechanical engineering, industrial equipment makers.
Version histories capture engineering judgement: what was changed, why, and what it cost. Models trained on them learn the reasoning, not only the final geometry.
Typical datasets
- 3D models and 2D production drawings
- Bills of materials and calculation notes
- Full version histories of the same project
- Specifications and technical review minutes
What we check first
- Client-owned designs and work under NDA are excluded unless the client consents.
- Export-controlled or defence-related material is out of scope.
- Supplier and pricing data are reviewed for commercial sensitivity.
Film & documentary
Production companies, documentary studios, streaming platforms with original content.
Professionally shot footage with its metadata is the scarcest input for video models. Rushes in particular are unique, never published, and rich in real-world variety.
Typical datasets
- Raw footage and unedited rushes
- Original productions with clear chain of title
- Scripts, transcripts and subtitles
- Shooting metadata and logs
What we check first
- Interviewees' image rights must cover the intended use, or the footage is excluded.
- Licensed music is stripped or the content excluded.
- Co-producer, broadcaster and author agreements are reviewed title by title.
Architecture & real estate
Architecture firms, surveyors, property developers, property managers, letting agencies.
Spatial models are learning to read and produce plans. Floor plans matched with photographs and written descriptions of the same space are what they need, and they are rarely published together.
Typical datasets
- Floor plans, BIM models (Revit, IFC) and building permit files
- Renderings and before / after project photography
- Property condition and inspection reports with photos
- Listing descriptions and property photography
What we check first
- Architects' moral rights and client ownership of commissioned plans are reviewed.
- Addresses, occupants and personal belongings in photos are removed.
- Listing content syndicated to portals may be subject to portal terms.
Manufacturing & quality
Plants, workshops and subcontractors in metalwork, plastics, food processing, electronics.
Industrial AI depends on labelled examples of what goes wrong. Years of defect images and quality reports are exactly the data that cannot be simulated convincingly.
Typical datasets
- Visual inspection images with defect labels
- Process parameters and machine sensor time series
- Non-conformity reports and corrective actions
- Work instructions and standard operating procedures
What we check first
- Customer part designs and specifications are excluded unless the customer consents.
- Trade secrets in process parameters are scoped with you before any sample.
- Operator names in reports are removed.
Field services & maintenance
HVAC, lifts, plumbing, electrical, facility management, equipment servicing.
A technician's report links a symptom, a diagnosis and a fix. Thousands of them teach models how physical systems fail and how they are repaired.
Typical datasets
- Intervention reports and diagnostic notes
- Before / after photographs of repairs
- Equipment histories and maintenance logs
- Parts lists and service quotes
What we check first
- Customer names, addresses and access codes are removed.
- Manufacturer documentation reproduced in reports may belong to the manufacturer.
- Photos of private premises are reviewed for identifying details.
Legal, accounting & consulting
Law firms, accounting firms, consultancies, engineering and audit practices.
Language models are being trained to do expert work. Professional writing in French, produced to a high standard, is scarce and in demand.
Typical datasets
- Templates, model clauses and standard documents
- Anonymised opinions, memos and reports
- Methodologies, checklists and training material
- Question and answer exchanges with clients, anonymised
What we check first
- Professional secrecy rules apply: only documents that can be fully anonymised are in scope.
- Client-specific deliverables require client consent or are excluded.
- Regulated professions confirm compliance with their professional body's rules.
Publishing, photography & audio
Publishers, press titles, photo agencies, podcast and music producers.
Curated, edited and captioned content is the reference standard for generative models. French-language catalogues are underrepresented in training data.
Typical datasets
- Book, magazine and press archives
- Photo libraries with captions and metadata
- Podcasts, radio archives and voice recordings
- Music stems and production sessions
What we check first
- Author, photographer and performer contracts must cover AI training.
- Collective management organisation mandates are reviewed.
- Voices and faces of identifiable people require a valid legal basis.
Software & IT
Software publishers, IT services companies, digital agencies.
Coding models improve most from real repositories where the reasoning is visible: the bug, the discussion, the fix and the review.
Typical datasets
- Proprietary source code with its commit history
- Tickets, bug reports and their resolutions
- Technical documentation and architecture decisions
- Code review discussions
What we check first
- Client-owned code is excluded unless the client consents.
- Secrets, keys and credentials are scanned for and removed.
- Open-source licences of third-party components are respected.
Customer support & sales
Contact centres, e-commerce, insurers, utilities, B2B sales teams.
Assistants and agents are trained on how real requests are resolved. Conversation histories in French, with their outcome, are directly usable.
Typical datasets
- Support tickets and chat transcripts
- Knowledge bases and response macros
- Product catalogues with photos and descriptions
- Sales proposals and objection handling scripts
What we check first
- Customer conversations are personal data: full anonymisation is mandatory.
- Call recordings require that customers were informed; otherwise they are excluded.
- Payment and health information is always excluded.
Agriculture & environment
Farms, cooperatives, agronomy consultants, landscaping and forestry companies.
Agricultural models need ground truth across seasons and regions. Multi-year records from real plots are hard to obtain at scale.
Typical datasets
- Field, crop and plant disease imagery
- Yield, soil and treatment records
- Drone and satellite surveys
- Agronomic advice and field reports
What we check first
- Land ownership and geolocation of farms are generalised where needed.
- Data shared with cooperatives or equipment makers may be subject to prior agreements.
- Drone imagery is reviewed for neighbouring properties and people.
Logistics & transport
Carriers, freight forwarders, warehouses, fleet operators.
Autonomy and planning models learn from real operations, including the exceptions. Incident logs and footage capture situations that are rare and valuable.
Typical datasets
- Delivery, route and incident logs
- Dashcam and warehouse camera footage
- Transport documents and customs files
- Load planning and warehouse layouts
What we check first
- Drivers and the public in footage are blurred or the footage excluded.
- Consignee names and addresses are removed.
- Employee monitoring rules under French labour law are respected.
Your sector is not listed?
Buyer demand evolves quickly. Tell us what your company holds and we will tell you plainly whether it is in demand.