Data for generative AI
Prompts, preferences and ratings with a written reason behind every judgement.
What Generative AI needs
Fine-tuning and alignment depend on human judgement that you can inspect. A preference score with no reason is hard to debug when the model drifts.
We produce the prompts, answers, comparisons and ratings your model needs, written and judged by qualified people against a versioned rubric, with the rationale kept alongside.
Data types
- Text
- Dialogue
- Code
Industries
- Technology and software
- Banking and finance
- Healthcare
- Education
- Retail and e-commerce
- Telecom
The data it takes
Instruction and response data
Prompts and high-quality answers across tasks, topics and difficulty levels.
Preference comparisons
Pairwise and ranked comparisons with a required rationale.
Reasoning traces
Step-by-step solutions checked for validity, not only the final answer.
Safety and red-team data
Adversarial probes and category-tagged results from vetted pools.
Multilingual generation data
Instructions and answers written natively in each language.
How we collect it
- 01
Taxonomy first
A topic, task and difficulty plan you approve, so coverage is designed.
- 02
Qualified writers and raters
People tested against the rubric before live work and re-certified monthly.
- 03
Rationale required
Every preference or rating carries a short written reason.
- 04
Slice-level agreement
Agreement is reported per slice so weak areas are visible.
Services that deliver it
Generative AI services
Prompts, preference pairs, rubric ratings and red teaming for language models.
Model evaluation
Human benchmarks that report rater agreement alongside the score.
Crowd-as-a-service
A vetted, managed workforce working inside your tools and your rules.
Translation and localisation
Parallel corpora, localised content and multilingual datasets, translated by native professionals.
Related datasets
Agent and customer chat
Support chat across industries with intent, entity and resolution labels.
Learn more
BFSI agent and customer chat
Banking support chat covering disputes, onboarding, cards and collections.
Learn more
Retail agent and customer chat
Order, returns, delivery and product enquiry conversations.
Learn moreTell us about your Generative AI project.
Describe the model and what it gets wrong today. You get volumes, languages, a quality target and a price, usually inside two working days.
- Reply within one working day
- A pilot batch you accept before we scale
Questions we get asked
Is any of the data machine-generated?
Only where your guidelines allow a model draft, and always reviewed and corrected by a person. We label which items had a draft.
Can you run restricted red teaming?
Yes, in vetted pools with access controls and NDA, with output delivered in a restricted export.
