Skip to content

The knowledge artificial intelligence has yet to read is inside companies.

Sourcing and qualifying knowledge bases for AI training.

Where data comes from.

What we have measured

+5,000,000
design files inventoried
50 million
lines of code with full version history
+120TB
de áudio
Legal, editorial and photographic
under assessment

Measured file by file, not estimated.

What has changed

  1. The open internet has already been read.

    AI models learned from what was published. What is still missing is kept inside companies and institutions: decades of work that was never cataloged or prepared for machine reading.

  2. Without provenance, the collection is unusable.

    Before considering volume, buyers want to know where the material came from, who owns it, and who has the right to license it. Data without a trail is a legal risk for the buyer and unusable material for the trainer.

  3. Provenance changes what a model learns.

    A model trained on data obtained without consent produces technology no one trusts. Trained on knowledge licensed with clarity, it comes closer to how people work and think.

What we do

The collection arrives scattered and leaves ready for buyer acceptance and verification.

  1. Find

    We reach the collection before it exists as a product and talk with people who have never thought of this material as an asset.

  2. Establish

    We resolve ownership, consent, and chain of custody before any discussion of value.

  3. Deliver

    We organize the raw collection to the format and standards the buyer requires, file by file.

Tracks

The types of collections we work with. Each arrives in its own formats and requires its own path to establish provenance.

Diagram: the knowledge base at the center, surrounded by five orbits. KNOWLEDGE BASE

  • Layered design

    .psd .ai .indd .fig .sketch .xd

    Identity and branding, campaigns, media, editorial, packaging, and interfaces.

  • Code

    git .js .ts .py .java .php .sql

    Web and mobile applications, backend, legacy systems, and technical documentation.

  • Audio and voice

    .wav .flac .mp3 .m4a .ogg

    Contact center audio, meetings, interviews, and content with multiple speakers.

  • Technical project

    .dwg .dxf .rvt .ifc .step .shp

    Architecture, electrical and electronic engineering, mechanical engineering, and geospatial data processing.

  • Legal

    Pleadings, rulings, legal opinions.

These are the tracks with collections already organized. Other verticals are under assessment.

The collection may be yours

It doesn't need to be cataloged, organized or in a current format. A company server, a drive nobody opens, a backup from an old project: that is almost always how we find it.

Extension, format and organization are our work. What needs to be clear first is where the material came from and who holds the right to license it.

Tell us about a collection

Why provenance

The next stage of artificial intelligence depends on material with an owner.

  • Every collection has a creator.

    Project files, contact center audio recordings, legacy system code, campaign assets. Each was created by someone, within a company, under some agreement.

  • The risk doesn't disappear.

    When material enters a model without known provenance, the problem passes to the buyer, the trainer, and ultimately the people who use the technology.

  • We resolve it first.

    We establish the provenance of every collection before it circulates. This means contracts, documents, and conversations, all handled before any sale.

Contact

If your company has a collection, get in touch by WhatsApp or email.