Every AI product is built on training data. The quality of your data determines the quality of your model.

Our standard accuracy rate is 99.1%, calculated per project as: (total items delivered minus errors identified in QA review) divided by total items delivered, multiplied by 100. This is not a self-reported marketing figure — it is calculated from the actual QA review data for each project and appears in the quality report attached to every delivery.

Before any project begins, we identify the most ambiguous annotation scenarios in your dataset and resolve them in writing with your team. The outcome is a documented decision tree included in the annotation guidelines. Every annotator follows the same decision tree for the same edge case — which is how you get consistency across a team of 50 annotators working on the same project.

Total items delivered, QA coverage percentage, number of items flagged, breakdown of flag types (boundary precision, class error, missed object, ambiguous annotation), number of corrections made, and the final accuracy rate. The report is delivered as a PDF alongside the annotated dataset.

Yes — and this is required, not optional. Every project begins with a guidelines document. If you already have annotation guidelines, we review them and propose improvements. If you do not, we draft them. Work does not start until you have approved the guidelines in writing.

You work with a named team of trained annotators and a named QA Lead, assigned to your project specifically. We do not rotate staff mid-project. The annotators who complete your pilot batch are the annotators who work your month-four delivery. Consistency of annotators produces consistency of labels.

Computer vision annotation is completed in CVAT (Computer Vision Annotation Tool), using bounding box, polygon, and segmentation tools natively. NLP and text annotation tasks are completed in Label Studio. Both platforms support standard export formats. We can work with client-provided tooling if required — contact us to discuss.

Our team of 50+ trained annotators can handle concurrent projects across computer vision and NLP. For a single large project, we can scale annotation throughput to 80,000+ images per month with maintained QA standards. Contact us to discuss dedicated team arrangements for enterprise volumes.

Your data is accessed only by the annotators assigned to your specific project — not shared across our team, not routed through third-party platforms, and not retained after delivery. We sign NDAs on request. If your project requires compartmentalised access where no single annotator can see the full dataset, we can implement that workflow.

Yes. Our standard NDA covers: confidentiality of dataset content and client identity, prohibition on secondary use of annotated data, data deletion on project completion, and restrictions on subcontracting without written consent. Contact us before project start to initiate the paperwork.

Via Google Drive, WeTransfer, or a client-provided S3 bucket. We confirm the method on project kick-off. We do not accept data via email or unencrypted channels. We confirm receipt within 2 hours and begin work within 24 hours of receiving the dataset and approved guidelines.

The free pilot covers up to 200 images or 500 text samples. It is annotated to the same standard as a paid project — same guidelines document, same dual-pass QA, same quality report on delivery. It exists so you can evaluate our quality on your actual data before committing to a paid project. No obligation to continue. Most clients who complete a pilot place a paid order within 7 days.

Book a scoping call or submit a pilot request today. We schedule scoping calls within 24 hours of a request. After the call, we send a project proposal and annotation guidelines within 48 hours. Work begins within 24 hours of receiving an approved guidelines document and your dataset.