Document AI
Global invoice extraction corpus
1.2M double-keyed pages across 19 countries with normalised tax fields, delivered as JSONL plus page-level confidence.
99.6% field accuracy
Portfolio
Selected work from our delivery floor.
Anonymised where clients require it, but the scope, volume and measured quality are exactly as shipped.
- Projects delivered
- 260+
- Records processed
- 180M+
- Languages covered
- 42
- Delivery centres
- 6
Case studies
Projects across every data modality
Computer vision
Retail shelf perception set
480,000 annotated shelf frames with SKU polygons, facing counts and out-of-stock flags for a grocery vision model.
0.92 mean IoU
Agents
Support agent evaluation harness
Authored 26,000 grounded instruction/response pairs and a rubric-scored eval suite for a deployed customer-support agent.
31% fewer escalations
Speech
Multilingual call transcription
3,100 hours of transcribed and diarised contact-centre audio across 11 languages with speaker and sentiment tags.
4.1% word error rate
Autonomy
Adverse-weather LiDAR sequences
310 hours of fused camera and LiDAR sequences focused on night, rain and fog edge cases for a robotics perception stack.
0.94 audit IoU
RLHF
Preference ranking programme
40,000 weekly pairwise comparisons from a calibrated rater panel with continuous agreement monitoring.
0.81 rater agreement
Partner network
A delivery network across four regions
We work with vetted data companies in China, Russia, the USA and the Middle East so language coverage, timezone coverage and surge capacity are never a bottleneck.
Shenzhen, China
Silk Data Works
Vision and OCR corpora for east-Asian scripts.
Hangzhou, China
Hanyu Vision Labs
Retail shelf and traffic-scene collection.
Moscow, Russia
Volga Datalytics
Cyrillic NLP corpora and speech transcription.
St. Petersburg, Russia
Neva Speech Corp
Multi-accent audio capture and diarisation.
Austin, USA
Meridian Data Group
Document extraction and fintech ground truth.
Seattle, USA
Northline Labeling
LiDAR cuboids and autonomy sequence review.
Dubai, UAE
Gulf Cognitive
Arabic annotation and bilingual QA pods.
Riyadh, Saudi Arabia
Levant Data Partners
Regulated-sector entry and compliance review.
Need labelled data by next sprint?
Send us a spec, a schema, or a handful of raw files. We return a pilot batch of 1,000 records with a quality report within five working days — free of charge.
info@betalen.in