Approach
How we work with clients
Engagements are organized from an agreed scope to documented delivery. The sequence below is how we typically run the work. It is adapted to the project at hand, and not every engagement uses every stage.
Engagement sequence
From scope to documented delivery
A project may begin with an existing collection, or may need only annotation and delivery. When the full path is required, it generally follows the order below.
-
01
Scope and rules
Before collection begins, the work is scoped: what will be gathered, what may be held, how it may be transformed, and what will be delivered. Those decisions are written down so later stages can be traced back to them, and so exclusions are as clear as inclusions.
-
02
Collection
Collection is carried out against that scope. The point is a set of sources and methods that can be accounted for, so that extraction and construction do not have to reconstruct decisions after the fact.
-
03
Extraction and governance
Collected material is reduced to usable content, and documented rules are applied to how that content may be stored, changed, and released. Transformations are recorded so they can be reviewed or repeated.
-
04
Construction
Depending on the engagement, this stage produces an LLM corpus, a multimodal dataset, a working data platform, or some combination of these. Construction is specified against the intended training, evaluation, or holding use.
-
05
Specialist annotation
Where the work requires the distinctions of a professional discipline, specialist annotation is organized against an agreed label space and guidelines, with review. This stage is used when the domain calls for it, not as a default added to every corpus.
-
06
Delivery
Delivery includes the data itself and a record of composition and construction — what went in, what was removed or transformed, and how the result is organized — so the work can be used and later revised with a clear record.
How engagements are run
Work is scoped before collection, or before the first stage that the engagement actually requires. Governance rules are written down and applied as part of the work, not added after a corpus has already been assembled.
Documentation travels with the data: a delivered corpus or dataset is accompanied by a description of its composition, the rules applied to it, and the limits that a later user should understand. Platforms and processing environments, when they are part of the engagement, are sized to the project rather than sold as a generic product stack.
The six offerings on the Services page are the substance of this work. This page describes the order and the working habits; that page describes each offering in more detail.
Who this is for
This kind of work is typically commissioned for programs such as the following. These are types of buyer situations, not accounts of named clients, and not a claim that particular organizations have engaged this company.
- Language-model training programs that need a specified corpus rather than an undifferentiated scrape.
- Evaluation work that depends on held-out or targeted test material whose construction can be described.
- Projects that must align additional modalities — such as image, audio, or document layout — with the text or labels they accompany.
- Professional-discipline programs that need specialist annotation and a delivered, documented corpus.
- Efforts that need a governed environment for holding and preparing data before a later training or evaluation stage.
- Revisions of an existing corpus, where later versions must be comparable with a clear account of what changed.
Start a conversation
If you are scoping a data or corpus engagement, start on the Contact page.