SCOPE
250–500 examples
On your exact task and rubric — code-reasoning preference pairs, coding-agent evals, STEM annotation or red-teaming.
Vetted Indian expert pools for code-reasoning preference data, coding-agent evals, STEM annotation and red-teaming. Contributors pass your rubric before production — every batch ships with seeded-gold QA and the scorecard.
250–500 EXPERT-LABELLED EXAMPLES · 5–10 BUSINESS DAYS · QA SCORECARD INCLUDED · FIXED-SCOPE PAID PILOT · REDO GUARANTEE
Send us one hard task and your rubric. We vet a matched expert pool, label 250–500 examples, and return the data with an agreement report and QA scorecard in 5–10 business days.
On your exact task and rubric — code-reasoning preference pairs, coding-agent evals, STEM annotation or red-teaming.
Task-specific contributor screening, seeded-gold QA in every batch, inter-annotator agreement report, founder review before delivery.
Code/STEM $2,500–$4,500 · Indic/multilingual $1,500–$2,500, by task complexity and review depth. No lock-in.
If the batch misses the QA threshold we agree before kickoff, we redo the affected work at our cost.
A live, scored 7-task screening — code review, STEM reasoning, RLHF preference on prose and code, classification, instruction-following and coding-agent eval rating — graded against a gold standard. Run it yourself in the browser.
Four workstreams, one standard of quality — scoped to your rubric, staffed by vetted specialists.
Code-generation review, coding-agent evaluation and STEM reasoning from screened engineers and PhDs — not a generic crowd.
We build rubrics, run them at scale, and surface the failure modes benchmarks miss.
Pairwise comparisons, rankings, and written rationales to train and stress-test reward models.
Native testers across Hindi, Tamil, Telugu, Bengali, Marathi, Kannada, Malayalam, Gujarati, Punjabi, Odia and more.
Most vendors ask you to trust their workforce. We give you the measurement layer: every pilot is judged against your task, your rubric, and the QA threshold we agree before kickoff.
We recruit for depth in the domains AI teams actually pay for.
IIT/NIT-grade software engineers for code generation, review, and coding-agent evaluations.
Maths, science and domain specialists for reasoning data and hard evals.
Fluency-tested native speakers for multilingual data across India's major languages.
You bring the task, rubric, and volume. We pressure-test the spec and set the gold standard.
We screen and NDA a scored expert pool matched to your domain and languages.
Production runs with seeded gold and multi-pass review — quality is measured, not assumed.
You get clean data plus agreement reports; we tighten the rubric and scale what works.
We're piloting expert annotation of professional-work video for embodied-AI and VLA training — causal event graphs, action-language captions, and 2D contact/keypoint labels, building on our own annotated multi-profession video work. Available for selected pilot discussions while our production focus stays on code/STEM evals and preference data.
We start behind an NDA, before any of your data changes hands.
We screen every contributor on your rubric — not a generic platform test — seed gold items into every batch, and you deal directly with the founders who own delivery. Smaller, more specialised, accountable per engagement.
Try the screening our annotators pass, live in your browser, and start with a small paid pilot judged on your own acceptance criteria. You see the scorecard and agreement report, not just labels.
Per-contributor NDAs and IP assignment, a DPA, an access-controlled workspace with no subcontracting, activity logging, and a retention/deletion policy. GDPR- and India-DPDP-aware; SOC 2 / ISO 27001 on the roadmap. NDA before any data exchange.
Software engineering and competitive coding, maths/STEM and PhD-level domains, native Indic languages, and — in pilot — physical-skill video for embodied AI.
Pilots run in 5–10 business days and are paid: code/STEM $2,500–$4,500, Indic/multilingual $1,500–$2,500, priced by task complexity and review depth. No lock-in. After a successful pilot we scope production batches and dedicated pools on a per-engagement basis.
Yes — every batch ships as labelled_data.json plus an inter-annotator agreement report and a gold scorecard (gold-pass rate, per-annotator accuracy, agreement, reviewer notes). View the redacted sample scorecard & agreement report — the labelled-data sample is downloadable from that page. Email pilots@goldset.ai for our security pack (NDA, DPA, data-flow and deletion policy).
Send us one task and a rubric. We'll vet a pool, label a sample, and return measured quality — data, agreement report and scorecard — in 5–10 business days.
SMALL · PAID · NO LOCK-IN
Join the vetted pool. Take the 7-task screening and, if you pass, get matched to paid AI-data work.