Human data · Latin America

Your model has never used the apps it will be asked to operate.

Pluri captures what real people see and do inside Latin America's apps: payments, checkout, delivery, public services, cancellation flows. Step by step, from their own screens, with logged consent and per-record provenance.

Not scraped. Not synthetic. Not reconstructed from documentation.

Send a spec Inspect one record
record ACCEPTED

    capture pipeline, live · illustrative values

    The gap

    Latin America is not missing from the internet. It is missing from the interfaces.

    54.7%of all retail payment transactions in Brazil run on Pix, a rail no crawler has ever seen · Banco Central do Brasil, H2 2025
    700M+native speakers of Spanish and Portuguese worldwide
    ~6%combined share of both languages in global web text corpora

    Web text gives a model Portuguese and Spanish. It does not give it the eleven screens between opening a banking app and confirming a transfer, or what a delivery checkout actually shows someone in Recife on a Friday night.

    Those surfaces are authenticated, personalized, ephemeral and closed to crawlers. They exist only on the device of the person using them, and only for as long as that person is looking at them.

    Pluri exists to collect them, at contract grade.

    Flows

    Step-level capture from inside the apps people actually use.

    Organised by rail, because that is how the flows behave and how models fail on them.

    01 Instant payments 02 Commerce and checkout 03 Delivery 04 Mobility 05 Government and public services 06 Subscriptions, paywalls and cancellation

    What ships

    • action traces, not screenshots · tap, field entry, wait, error, retry, abandon
    • instruction compliance verified per step
    • redaction at ingestion, logged per record
    • targeting by country, city, device, OS and profile

    How it is scoped

    • your spec, our field · costed pilot in 72h
    • daily batch close with manifest
    • your verdict template, reconciled per unit
    • rejected units replaced, acceptance rate reported

    Including the flows that fail.

    Agents do not break on the happy path. They break on the expired token, the rejected CPF, the timeout at step nine. We collect those on purpose.

    Panels

    The same machine, left running.

    Consented cohorts of 500 to 2,000 people per sector submit screen-level evidence every month from their own devices. Same people, wave after wave, so a movement is a movement, not sampling noise.

    REVENUE 01

    Syndicated subscription

    Monthly delivery combining the record-level dataset and an aggregated read: price and promo movements, share shifts, flow changes. Sold to operators in the sector.

    • monthly drop · dataset plus aggregated read
    • same cohort tracked across waves
    • every record ships with full provenance
    REVENUE 02

    Training licence

    The same underlying screens, licensed for model training. One collection, two revenue lines, because the panel consent covers both uses explicitly from day one.

    • licence on the underlying corpus
    • consent versioned for both uses
    • non-exclusive by default

    No client-targeted collection. Ever.

    No client can commission capture aimed at a named competitor. This is not a courtesy, it is structural protection: it settles antitrust exposure, platform relationships and conflict between subscribers with a single clause. Everyone receives the same deliverable, so no one is buying surveillance of anyone.

    Proven in production: our standing screen-collection programme runs at a contracted volume of 20,000 validated units per month.

    One record

    Every claim about quality reduces to one question: show me a record.

    This is the audit trail of a single delivered unit, field by field, the way it ships.

    record plr_a91c04…7e4 APP FLOW · PIX · PT-BR ACCEPTED

    Capture

    rail
    instant payment · transfer to new payee
    steps
    11 screens · open to confirmation
    outcome
    completed (1 retry at step 7, captured)
    device
    Android 14 · 1080x2400 · pt-BR locale

    Contributor

    id
    c_7f20…9b (pseudonymous)
    profile
    woman · 25 to 34 · Goiânia · own account
    identity
    name, phone and payout identity on file, checked at payout
    history
    61 accepted units · 0 fraud flags

    Consent

    version
    v2.1 · accepted 2026-07-14 09:12 BRT
    scope
    commercial model training · worldwide · sublicensable
    withdrawal
    honoured forward, logged

    Machine QA

    signals read
    11 per screen (amount, timestamp, state, target, completion)
    dedup
    pHash 9e1a7252… · no match in full history
    redaction
    3 regions masked · log retained per screen

    Human gate

    review
    approved · reviewer r_04 · 2026-07-15
    client verdict
    accepted · reconciled per unit

    Settlement

    payout
    Pix · released on acceptance, never before
    counterparty
    registered company, tax ID on contract

    Ask any vendor for this view of one record. That question is the entire audit.

    Schema as written by the production pipeline on every unit we ship. Values anonymised.

    The machine
    01Reach. Owned, identified base. No marketplace, no anonymous crowd.
    02Execute. Guided capture that validates at submission time.
    03Verify. Machine QA, then a human gate on every batch.
    04Pay. Local instant rails, released on acceptance.

    Also collected

    Speech, imagery and documents.

    The same panel and the same controls, applied to conventional modalities. Speech with enforced accent quotas across pt-BR and es-419. Imagery, receipts and shelf capture. Available as add-on scope or standalone, when the buyer already knows what they need.

    48kHz WAV or client spec · demographic attributes per contributor · perceptual-hash dedup · same provenance record

    Engagements

    Three ways to work with us.

    One collection machine, three commercial shapes. Start with a scoped pilot, or contract capacity from day one.

    For AI labs and agent teams

    Custom collection

    Your spec, our field.

    Send the rail, the volume and the geography. We come back with scope, timeline and unit pricing, and run the collection under your QA bar.

    • costed pilot in 72 hours
    • daily batch close with manifest
    • your verdict template, reconciled per unit
    • rejected units replaced, acceptance rate reported

    For operators in-region

    Syndicated panel

    What the market actually shows people.

    A monthly subscription to a standing cohort in your sector, whether fintech, retail, delivery or media: prices paid, offers seen, flows completed, wave after wave, from the same people.

    • monthly drop · dataset plus aggregated read
    • same cohort tracked across waves
    • no collection aimed at a named competitor
    • annual term, sector exclusivity never sold

    For data platforms

    Reserved capacity

    Supply you can plan against.

    Contract a monthly throughput floor and stop treating Latin American supply as spot-market risk. Burst on demand when a client lands.

    • monthly throughput floor, by contract
    • priority quotas by country, city and profile
    • burst capacity on demand
    • quarterly quality review with acceptance data
    1.4M+identified people in the owned panel
    20+cities with active field operations
    20kvalidated units per month, under contract
    2019running field data operations since

    Start

    Send a spec. Get a costed pilot in 72 hours.

    One page is enough: rail, volume, country and city targets, and your QA bar. We reply with scope, timeline and unit pricing.

    hello@pluri.lat typical first pilot: one city, one rail, two to four weeks