What we can take on
2D bounding boxes
Detection labels at volume across images and video frames.
Polygon and mask
Instance and semantic segmentation for irregular shapes.
Keypoint and pose
Skeletal points for pose, gesture and facial landmark work.
3D cuboids and LiDAR
Point cloud annotation with camera and LiDAR synchronised.
OCR and document
Word boxes, key-value linking, table structure and handwriting.
Text labelling
Entities, intents, sentiment and custom taxonomies.
The types of data annotation we run
The full breakdown, not the headline categories. Anything not on this list is a scoping call away.
Text annotation. Sentiment and emotion tagging, Named-entity recognition, Intent and taxonomy classification, Summarisation and extraction review, Question-answer pair validation, Toxicity and safety flagging.
Audio annotation. Speech transcription verification, Speaker identification and diarisation, Sound-event tagging, Language and dialect identification, Emotion and tone tagging.
Image annotation. 2D bounding boxes, Polygon and semantic segmentation, Keypoint and pose estimation, Facial landmark annotation, OCR word and layout boxes, Image-level classification.
Video annotation. Multi-object tracking, Action and activity recognition, Instance segmentation across frames, Lane and drivable-area marking, Frame-level event tagging, Temporal boundary annotation.
3D and sensor annotation. LiDAR point-cloud cuboids, Camera-LiDAR sensor fusion, Radar and depth-map labelling, Spatial scene segmentation.
The forms the data takes
What you actually receive: file types, structure and the metadata that travels with it.
- Image: boxes, polygons, masks
- COCO, Pascal VOC and YOLO exports, with class IDs from your taxonomy.
- Video: tracking and events
- Per-frame tracks with persistent object IDs, interpolation and event boundaries.
- Text: entities, intent, sentiment
- Span and document labels in JSONL, CoNLL or your own schema.
- Audio: segments and tags
- Timestamped segments, speaker and event labels, in JSON or TextGrid.
- 3D and sensor
- Cuboids and point-cloud segmentation for LiDAR and depth data.
- Documents
- Word boxes, key-value links and table structure for OCR and document AI.
How the data is collected
From the first brief to the batch you receive, step by step.
- 01
You send the raw data, or we collect it
Annotation starts from your files through a secure transfer, or from data we collected for you, so the label and the source stay linked.
- 02
Guidelines with worked examples
We turn your labelling spec into written guidelines with examples of the edge cases, and agree them with you before volume work starts.
- 03
Pre-labelling where it helps
Model or heuristic pre-labels are used only where reviewers can fix them faster than they can draw from scratch, and never to skip review.
- 04
Layered review
Peer review, then expert review on a risk-weighted sample, with gold tasks injected so label quality is measured while the work runs.
- 05
Disagreements settled, not averaged
Contested items go to a senior reviewer, and the ruling is written back into the guidelines.
How a project runs
- 01
Scoping call
We agree volume, languages, demographic quotas, formats and what counts as a rejected item.
- 02
Guidelines and gold set
Written guidelines with worked examples, plus a gold set every contributor is tested against.
- 03
Pilot batch
A small paid batch you accept before anything scales. Most guideline problems surface here.
- 04
Production
Work flows through capture checks, automated QC, peer review and expert review.
- 05
Delivery
Incremental batches in your format with a data card, QA report and agreement scores.
- 06
Iteration
Rejections feed back into the guidelines so the same mistake is not made twice.
How quality is checked
Checks that run during the work, not a sample at the end.
- Inter-annotator agreement reported per class
- Gold tasks injected into the queue at a fixed rate
- Per-class IoU scoring for visual labels
- Reviewer sampling raised automatically on low-agreement classes
- Rejection breakdown delivered with every batch
Industries that use it
Where teams bring us data annotation, and what they use it for.
- AutomotiveLane, sign, pedestrian and in-cabin labelling.
- Retail and e-commerceProduct, shelf and queue annotation.
- HealthcareClinician-supervised labelling of notes and images.
- Banking and financeDocument entity and field extraction.
- Safety and surveillanceActivity and anomaly labelling.
- Real estateFloor-plan parsing and property image tagging.
- AR and VRHand, gesture and scene understanding labels.
- ManufacturingDefect and part-level visual inspection labels.
Why teams pick us for data annotation
IoU against gold
Accepted work clears 0.9 IoU on the gold set, per class, not on average.
Consensus on hard classes
Ambiguous classes are labelled two or three times and reconciled.
Class-level reporting
You see which classes your annotators disagreed on, not one headline number.
Secure workspaces
Private NDA-only pools, secure view mode and no downloads where required.
