What we do
Human data your AI team can defend.
Sonexis helps AI teams source, assess and obtain appropriately licensed human data for multilingual production systems, starting with how India actually speaks. We check what already exists and can lawfully be licensed before anyone funds new collection.
Three routes, in this order
Every requirement is checked against them in sequence. Collection is the fallback, not the starting point.
Route 1
Existing reviewed supply
Where data already exists and has been reviewed for rights, provenance, metadata and quality, that is the fastest and cheapest route for a buyer. Any candidate dataset is located and reviewed against the requirement before a match is claimed. Sonexis does not operate an open public catalogue, and does not claim a match before one is found.
The fastest route when reviewed supply fits the brief. No collection cost, and the evidence is already in hand.
Route 2
Managed sourcing: locating and reviewing potential suppliers
Where suitable data may exist outside anything Sonexis has reviewed, we look for a supplier who holds it and review their identity, rights and provenance before anything is licensed. No supplier is described as reviewed until that work is complete, and the rights-holder for each dataset is identified per transaction.
Wider reach, and adds review time. Each requirement starts its own search, so the rights position is established for that dataset rather than inherited from a listing.
Route 3
Buyer-funded custom collection
When suitable existing data does not meet the requirement, Sonexis scopes and manages a funded collection directly or through approved collection partners, with defined consent, QA, metadata and delivery requirements. It starts once the requirement is funded, not before.
Meets briefs existing supply cannot. Longer and higher cost, which is why it runs only after routes 1 and 2 have been tested against the requirement.
Across all three
Evaluation is not a fourth route. It runs across all three: candidate data is assessed against the actual AI use case — the pipeline stage, the languages and code-switching involved, and the speaker structure the deployment needs — before it is offered, sourced further, or used to justify funding a collection. Suitability is never one score, and a set that fits recognition is not assumed to fit synthesis or agent evaluation.
Two boundaries a buyer should know about
Both exist to keep what a buyer is told about a dataset the same thing that is true of it.
Supplier-owned data stays supplier-owned
Where a dataset comes from a supplier, the source and rights-holder are named as such. Sonexis does not present supplier-owned or supplier-collected data as collected or owned by Sonexis, and a buyer licence never exceeds what the rights-holder actually holds.
Collection begins when it is funded
Route 3 starts once a requirement is funded. Sonexis does not run speculative collection against anticipated demand, which is what keeps a collection scoped to the brief that pays for it rather than to a guess about the market.
Custom collection, when it is the right route
Route 3 is a real Sonexis capability. Collection runs either directly or through approved collection partners working to a Sonexis specification, with the same consent, QA, metadata and delivery requirements applied in both cases.
Collection under Sonexis specification is recorded as linear PCM WAV. Sample rate and bit depth are confirmed per project against the buyer specification, and the delivered technical properties are stated per file rather than claimed in general.
Consent is captured against a versioned consent instrument. The scopes granted at the moment of submission are snapshotted onto the recording, so a later change to the consent text cannot retroactively widen what a past contributor agreed to. Export is gated against those snapshotted scopes.
See a Dataset PassportTell us what your model needs
Describe the requirement and we check whether suitable existing data exists and can be licensed, then run managed sourcing, and scope funded collection only where existing data cannot meet the brief.