A European computer vision company
Training data annotation from Uzbekistan
Volume annotation with guidelines calibrated before volume and inter-annotator agreement measured throughout.
Case facts
- Delivery model
- Global network
- Variant
- en
- Team
- 31 annotators, Tashkent
- Time to launch
- 3 weeks of calibration
The problem
Earlier projects started annotating before guidelines stabilised, and quality only surfaced as a problem in model performance months later.
What we built
A small initial batch annotated in parallel by several annotators, with disagreements discussed and guidelines rewritten until agreement stabilised. Only then did volume begin.
Compliance
Outside the EEA. Standard contractual clauses and a transfer impact assessment. Data is not used to train other models, by contractual clause.
Results
- inter-annotator agreement
- 0,89inter-annotator agreement
- calibration before volume
- 3 sem.calibration before volume
- annotators in production
- 31annotators in production
- cost index against Lisbon at 100
- 35cost index against Lisbon at 100
They insisted on three weeks of calibration before annotating anything. They were right.
Let us look at the numbers for your case
Tell us which processes you want to outsource, in which languages and at what volume. We come back with a euro estimate and an operating design, with no commitment.
We reply within 6 hours on working days. If you would rather write: info@corpshore.solutions