Who we are, what we collect, what we sell.
What we are
- A training-data company — Zyno AI Inc, backed by Y Combinator.
- A bench of 650 transcribers, aligners and reviewers on Foundry, our own annotation platform.
- Independent: no frontier lab owns a piece of us or sits on the cap table.
- SOC 2, ISO 27001, ISO 27701 and TPN Gold; GDPR, CCPA and HIPAA compliant.
What we collect
- Speech & audio4,000+ hours, 9 localesDelivered
- VideoTask video, egocentric, depthOpen
- ImageProducts, scenes, documentsOpen
- Text & conversationDialogues, prompts, transcriptsOpen
Plus embodied data for robotics — egocentric, depth, teleoperation, gripper — in early access.
Data collectionWhat we sell
- Custom collectionsper usable or finished hour
- Annotation on Foundryper audio hour, per label, per judgment
- Licensed datasetsper set, non-exclusive by default
- Benchmark runsper track, per sector
- RL environmentsper sector — design partners
- Physical-AI pilotsscoped per program
A working bench, not a landing page.
Every number here is speech work already shipped. When we take on a new modality it will be listed the same way — after it ships, not before.
- 4,000+
- Hours of audio delivered
- 650
- Transcribers and aligners on the bench
- 50 hrs
- Delivered every working day
- 95%+
- QA-sample accuracy, checked against WER
50 on fixed-term contracts, the rest freelance — and hiring against every open sector.
Freelance bench, timestamped to the word so a model can be scored on when it heard something, not just what.
Collect it, annotate it, or ask for the set that already exists.
Data collection
Speech delivered at scale. Video, image and text collected to spec.
Speech delivered · video, image, text openData annotation
The Foundry bench — transcription, alignment and labels, any data type.
Audio delivered · image, video, text on the same benchRequest data
Name the input your model fails on. We collect it, transcribe or label it, and license the set with the spec and terms published.
Priced by the unit it is delivered in
Four steps, and the first one is a phone call.
No dataset catalogue to guess from. You describe the failure; we scope, pilot and deliver against it, with the contributor rate on the quote.
Full delivery spec and licence terms- 01Tell us the failureday 0
Not the dataset — the input your model gets wrong. A clip, a photo, a paragraph is enough to start.
- 02Written spec and quote2–3 days
Quantity, locales, devices, conditions, annotation depth — and the contributor rate you will be paying, stated as its own line.
- 03Pilot batch1–2 weeks
Five percent of the volume, or ten to thirty episodes for embodied data, delivered first with a quality report. If the spec was wrong, the re-run is on us.
- 04Weekly deliveryongoing
Drops into your bucket in the region you chose, with the manifest, consent references and reviewer chain on every item.
Embodied data, collected by people who do the task.
Early access · first programs being scopedEgocentric video, depth, teleoperation and handheld-gripper demonstrations on the same recruiting, consent and QA machinery as our speech work. No volumes listed until the first collections ship.
This is a delivery, not a description of one.
Lowercase, fillers kept, one written convention for every tag, and a timestamp on every word — so accuracy and latency can be scored apart.
Read the methodhaan so i i filed the claim on monday [pause] but the portal is still showing pending only
let me pull that up for you ma'am [keyboard] can you confirm the claim reference
it's uh [pause] c l four two nine [overlap] eight—
[overlap] four two nine eight, yes i have it here
- Everything lowercase, punctuation only where it changes meaning.
- Fillers, repeats and false starts stay — they are what the model has to survive.
- [pause], [overlap], [keyboard] and other tags follow one written convention.
- Every word carries a start and end timestamp, so latency can be scored separately from accuracy.
Operators, not a crowd.
The bench is built from people who worked the floor, the ward and the claims desk — ex-contact-centre QA leads, claims handlers, ward clerks. They are paid by the unit, at rates published on the listing.
Two crafts, one bench. Transcribers write what was actually said, fillers and all. Aligners put a start and end time on every word. A rubric author is someone who signs off on these calls at work, and every rubric has to be applicable by a stranger.
Built for a procurement review, not just a demo.
- Consent captured on the file, before anyone speaks or films, naming what the data trains.
- Reviewers stream a clip; they never download it. Access is role-scoped and logged.
- Every collection is pinned to one region — Frankfurt, Northern Virginia or Mumbai — and stays there.
9 locales live. The next one is a recruiting problem, not a research one.
Contributors record on their own phones, in their own rooms, in the accent they speak. Blue is where the bench works today; anything else we recruit to order against locale quotas.
The protocol is public before the scores are.
We would rather publish how a run works and no numbers than numbers nobody measured. The design is fixed and public now.
Benchmarks
Two tracks — transcription and live conversation — run across ten sectors on held-out audio, scored against the person who does that job.
Protocol published · first run openEnvironments
Live speech-to-speech calls with trained speakers who interrupt, change their mind and hand over the phone mid-call.
Design partnersGet paid to talk. Twenty minutes, your own phone.
Anyone can join the recording pool. Pick a session, speak in your own accent, and get paid once it passes review — no studio, no interview, no shift to show up for.
- $20
- Base rate per approved audio hour
- 20 min
- Typical session length
- 7 days
- From approval to payout
- $10
- Minimum balance to cash out