AI Lab
The data factoryfor real-world AI.
Raw data goes in: captured with consent, labelled by native annotators and checked by people. Training-ready data comes out, with the models and agents built on it.
The data factory
Raw data in. Training-ready data out.
Real work, recorded with consent.
Data starts where the work happens. Domain datasets, field collection and enterprise systems come in alongside first-person video captured through KLKT, on a phone, a Roboband or a wristband.
- 67,795+ people on KLKT, in 20+ countries
- Consent in Arabic, English, Filipino and Urdu
Station 01 · CaptureSupported data types
Every data type, labelled by people.
Text
Entities, sentiment, moderation and preference labels, in Arabic and English.
What we label
- Named entities
- Sentiment
- Content moderation
- Ranked model responses
Sample recordIllustrative, not customer data
{
"text": "افتتح المعرض في دبي يوم الأحد",
"entities": [
{ "span": "دبي", "type": "LOCATION" },
{ "span": "يوم الأحد", "type": "DATE" }
],
"sentiment": "neutral",
"reviewed_by": "native annotator"
}What the data powers
Data, then models, then agents.
- 01 · Data
AI Lab data
Labelled at 98% accuracy.
Browse the Data Hub - 02 · Models
Munsit
#1 Arabic speech recognition.
Explore Munsit - 03 · Agents
Agentix
Arabic agents on your phone lines.
Explore Agentix
Case studies
Results already delivered.
- Financial servicesContract compliance review85%faster reviews60%faster application processingRead case study
- InsuranceClaims checked as they are submitted72%estimated cut in rejections60%less processing timeRead case study
- Real estateReal estate operations22%faster reviews28%less legal workloadRead case study
- Physical AIVision data for humanoid robots1.8Mframes annotated6monthsRead case study
- HRHiring across 3,000+ CVs32%faster hiring and onboarding21%lower admin costSpeak to an expert
- Supply chainSupply chain processing23%faster processing17%fewer data-entry errors (projected)Speak to an expert
How we deliver
Four phases to sign-off.
Research
What we publish.
Can we use one layer on its own?
Yes. Data services, Munsit and Agentix are each available separately.
What does the data layer cover?
Sourcing and collection, labelling of text, image, video and audio at 98% accuracy with human-in-the-loop quality checks, and physical-AI data recorded from real work.
How does a project start?
With discovery: the problem, the systems it touches and the data behind it. That becomes a roadmap, then a first build phase, a review, a second build phase and sign-off.
Can you deploy on our own infrastructure?
Yes. We deploy on-premises, hybrid or in the cloud, and hosting and processing can be kept in region.
