Request the datayour model actually needs.
Browse the kinds of data CNTXT captures and labels. Request a category or a custom collection.
Data captured from real work, labelled and checked by people
- Audio
Arabic conversational speech
Recorded conversations across Arabic dialects, transcribed and checked by native Arabic annotators.
- Video
First-person video of service work
Hands-free recordings of kitchen work, housekeeping, laundry, cleaning and hospital portering, each confirmed by human review.
- Video
First-person video of warehouse work
Head- and wrist-camera recordings of picking, packing and shelf stocking, made by consenting workers for robotics training.
- Document
Handwritten Arabic pages
Photographs of handwritten Arabic pages, submitted through the KLKT app and accepted only after human review.
VideoAction-labelled video
Video labelled frame by frame with the actions it shows, including first-person recordings of real work.
- Action labels
- Frame by frame
- Image
Annotated images
Images labelled with bounding boxes and segmentation masks, with human-in-the-loop quality assurance at 98% labelling accuracy.
- Bounding boxes
- Segmentation
- 3D
Annotated LiDAR
LiDAR scans annotated for perception models, with human review before delivery.
- LiDAR
- Human review
- Text
Content moderation labels
Text reviewed and labelled against your moderation policy, in Arabic and English, with human quality assurance.
- Moderation
- Policy review
- Evaluation
Expert evaluation sets
Test sets written and graded by domain experts in fields such as finance, healthcare, logistics and government.
- Domain experts
- Model evaluation
- Text
Ranked model responses
Prompts annotated and model responses ranked by people, for fine-tuning and evaluation.
- Eval ranking
- Prompt annotation
- Text
Sentiment and entity labels
Text labelled for sentiment and named entities, with human-in-the-loop quality assurance at 98% labelling accuracy.
- Sentiment
- Entities
- Audio
Transcribed audio
Audio transcribed, timestamped and split by speaker. Native Arabic annotators cover more than 25 dialects.
- Transcription
- Diarisation
How do I request data?
Add the categories you need to your request, then send it. The contact form opens with your list filled in. You can also ask for a custom collection.
Can you collect data that is not listed?
Yes. We source domain-specific datasets, collect in the field, integrate with enterprise systems and generate synthetic data in controlled environments.
How accurate is the labelling?
Text, image, video and audio are labelled at 98% accuracy, with human-in-the-loop quality assurance.
Which Arabic dialects do your annotators cover?
More than 25, including Gulf, Egyptian, Levantine and Maghrebi Arabic.
Need data that is not in the catalogue?
Tell us the task, the environment and the labels. We will scope a custom collection with you.
