Human data from people who are actually employed
Annotation, model evaluation and human feedback data delivered by degree-qualified teams on formal contracts — a workforce model you can describe publicly without a comms review.
Book a 20-minute callWe work directly with AI teams and as a delivery partner to data platforms.
What we deliver
Annotation
Text classification, entity tagging, sentiment and intent labelling, image and audio annotation, and structured enrichment — to your schema and guidelines.
Model output review
Side-by-side comparison, preference ranking and rubric-based scoring for helpfulness, accuracy, tone and instruction-following.
Human feedback data
Written demonstrations, ideal-response drafting and correction of model outputs — work that lives or dies on the writing quality of the person doing it.
Prompt and dataset building
Prompt writing to specification, adversarial and edge-case generation, and dataset construction against a defined taxonomy.
Search and relevance
Query-result relevance judgements and ranking evaluation against detailed rating guidelines.
Transcription and localisation QA
Audio transcription, translation review and localisation quality checks.
Dataset review
Independent quality assessment of existing datasets, error rate measurement, and re-annotation of labelled data.
Surge support
Additional throughput for platforms who've won more work than they can currently staff, without compromising on who does it.
Most data work fails on the same three things
Inconsistency between annotators, weak written English, and an inability to absorb guideline changes mid-project. We recruit and train directly against those failure modes.
What our annotators bring
Not a general labour pool with a task interface — a specified, trained and measured workforce.
- Degree-qualified, every oneUniversity graduates with excellent written and spoken English, recruited against a defined standard rather than availability.
- Subject-matter matchingWhere a project needs STEM, finance, legal, medical or software backgrounds, we recruit and assemble against that requirement rather than assigning whoever is free.
- Trained on guideline adherenceAcademy training covers working to written specifications, handling ambiguity, and escalating edge cases instead of guessing.
- Consistency, measuredGold-standard tasks, inter-annotator agreement checks and calibration sessions, reported against agreed thresholds.
- Guideline changes absorbed properlySpecs evolve. We re-brief, re-calibrate and flag where earlier work needs revisiting rather than quietly carrying on.
- Dedicated teamsAnnotators stay on your project, so accumulated context isn't lost between batches.
A supply chain you can stand behind publicly
How annotation work gets staffed has become a procurement question, not just an ethical one. Ours is straightforward.
Directly employed
Our people are employees on formal contracts — not task-based gig workers paid per unit.
Salaried, with progression
Pay is structured with defined routes into senior and specialist roles rather than piece rates.
Real benefits
Family healthcare, education support for employees' children, continuous training, staff transport and meals.
Open during procurement
We'll discuss pay structures, conditions and welfare provisions openly. Ask — and be wary of any vendor who deflects.
If your reputation depends on how the data was made, ask how the people were treated.
We'd rather that question came early than after publication.Where we stand today
Managed office environments where a project requires it, NDAs across all project staff, per-project access controls and device policies, and documented data handling, retention and destruction procedures agreed before work begins.
Formal certification is on our roadmap. We'll tell you precisely what we hold rather than implying more — ask on the call and you'll get a specific answer, which is a better basis for a decision than a badge on a webpage.
Prove it on a paid pilot
Nobody should award volume to an unproven vendor. Measure our output against your own quality bar first.
Scoping
Task type, volumes, quality thresholds, turnaround and any specialist background required.
Paid pilot batch
A sample you measure against your own bar before committing to anything larger.
Calibration
We work through disagreements and edge cases with your team until agreement rates hold.
Ramp
Volume increases in controlled stages with quality reported throughout.
Steady state
Ongoing delivery against agreed reporting, plus surge capacity when you need it.
Questions worth asking first
Can you staff projects needing specific expertise?
Yes. Where a project requires particular academic or professional backgrounds, we recruit against that specification and build a dedicated team for it rather than reassigning generalists.
How do you measure annotation quality?
Gold-standard tasks, inter-annotator agreement and audit sampling, with thresholds agreed during scoping and reported against throughout the engagement.
How quickly can you scale a team?
It depends on the specialism. General annotation teams ramp faster than specialist ones. You'll get a realistic answer during scoping rather than an optimistic one.
Will you run a paid pilot?
Yes, and we'd encourage it. A pilot measured against your own bar is a better basis for a decision than any claim we could make here.
Can you work as a subcontracted delivery partner?
Yes. We work both directly with AI teams and as a delivery partner to data platforms managing their own client relationships.
Do you take content moderation work?
We assess it case by case. Where it involves exposure to distressing material, we'll discuss the welfare provisions we consider necessary before agreeing — and we may decline work we don't believe we can support our people through.
What certifications do you hold?
We'll be precise about this on the call rather than implying more than we hold. If your procurement requires certification we don't yet have, that's worth establishing in minute five rather than month four.
How do you handle guideline changes mid-project?
Re-brief, re-calibrate, and flag where completed work may need revisiting. Quietly continuing under an outdated spec is how datasets get contaminated.
Send us a pilot batch
Share your task types, volumes and quality requirements, and we'll tell you honestly what we can deliver and when — then prove it.