Insights
Notes from the delivery floor
What we learn running annotation, validation and human-in-the-loop operations at production scale. Written by the people doing the work, not the people selling it.
What is human in the loop, and when does AI need it?
You have probably been a human in the loop without knowing it. Here is what it means, and the four signs your AI needs one.
Read the article →
Data labeling vs data annotation: is there a difference?
A label is a name tag. An annotation is a map. Why the difference decides whether an AI model learns the right lesson, or a shortcut.
Read the article →
Your agreement score cannot see half your errors
Inter-annotator agreement measures whether your labellers disagree. Most of the expensive damage comes from the times they agree, all in the same wrong direction.
Read the article →
The real price of getting AI data wrong
A data error does not cost what it costs to fix. It costs what it costs to fix, multiplied by how long it was allowed to live. Why detection speed beats accuracy rate as a quality metric.
Read the article →
Stay close to the work
Get the next note from the delivery floor.
One email when we publish something new. No newsletter filler, no sales sequence.
Talk to the delivery team
Send us a sample batch. We will send back the agreement score with it.
Managed annotation, validation and human-in-the-loop operations from Nairobi, ISO 27001 certified, with named delivery leads on every project.
Request a free trial batch