Xenveo collects, annotates, and evaluates training data for AI systems that have to work outside English, outside the cloud, and outside the assumptions a model picks up from the open web. Real people on the ground in eight countries. Native depth in six languages. Specialist review across regulated domains.
Most AI training data is scraped, licensed, or generated. Some kinds of data can't be — they have to be captured by people in physical places, doing real things, with cameras and microphones we put in their hands. That's the work that gets done here.
Image, video, audio, and egocentric (first-person) video. Structured capture protocols with consent, metadata, and quality validation built into every submission.
Active field operations across eight countries: the United States, Canada, the United Kingdom, France, Spain, the United Arab Emirates, India, and Brazil. Capture is run through trained local contributors with on-the-ground coordinators in each region.
Asset library exceeds one million captured items. Active capacity supports concurrent campaigns in multiple geographies; campaign-level throughput scales to the project.
Most multilingual annotation pipelines run through English. Guidelines are written in English and translated. Edge cases are escalated to English-speaking reviewers. Quality is measured against English-language baselines. The result is a model that performs in non-English languages the way a tourist speaks them — fluent enough to be understood, wrong enough that a native speaker can tell.
We annotate in the language the work was created in. Tier-one languages have full operational depth — native annotators, native reviewers, native escalation — and tier-two languages run on the same backbone with smaller pools.
Native annotators, native reviewers, native escalation. These are the languages we run end-to-end with no English bridge in the pipeline.
Same backbone, smaller pools. Additional languages opened on request when a project's scope justifies it.
Some judgments cannot be crowdsourced. A licensed pharmacist evaluating an AI's billing-code decision, a corporate attorney reviewing the tone of a generated NDA, a clinical researcher assessing diagnostic plausibility — these are not edge cases that a generalist reviewer can adjudicate with better instructions. They are the work, and the people doing them have to be qualified to do it.
Our reviewer pool exceeds one thousand vetted experts, organized by domain, geography, and language. Vetting includes credential verification, reference checks, calibration tasks, and ongoing quality scoring. Buyers see the credentialed pool that fits their project; reviewers see only the projects they are qualified for.
Practicing attorneys reviewing AI-generated drafts for accuracy, tone, and jurisdictional fit.
Licensed clinicians and pharmacists reviewing AI decisions in clinical, billing, and adjacent contexts.
Working engineers reviewing AI-generated code for correctness, security, and idiomatic style.
Credentialed accountants and financial professionals reviewing AI outputs in regulated contexts.
Domain researchers, working in collaboration with buyer teams, evaluating AI outputs against domain-specific quality rubrics.
Pattern:heavy multimodal capture across geographies, calibrated annotation in non-English languages, expert evaluation on outputs the model wasn't pretrained for. Engagements typically begin with a defined-scope capture pilot, expand into multilingual annotation programs as the model's coverage grows, and stabilize into long-running evaluation work once the model is in deployment.
Pattern:native-language annotation as the primary deliverable, with capture in-region for cultural and contextual grounding, and credentialed review for outputs in regulated domains (legal, healthcare, government). Engagements often run alongside the partner's existing systems-integrator relationship, with Xenveo positioned as a specialist sub-layer rather than a general vendor.
Pattern:lighter on capture, heavy on credentialed expert review. The work tends to be evaluation-shaped — judging an AI's output quality against domain rubrics — rather than data-collection-shaped. Engagements often begin as one-off audits and expand into ongoing review programs once the deployment scales.
Most engagements begin with a 4-to-6-week pilot — defined scope, agreed-on deliverables, written quality criteria. Field-capture pilots involving on-the-ground operations may run longer to accommodate sourcing, training, and on-site coordination. Pilots are designed to give both sides a real basis to scale from, not to be promotional.
We support regional data-residency commitments for buyers whose contracts or regulatory environment require it. Capture and annotation work performed in a region can be retained in that region throughout the engagement; cross-region transfer happens only with documented buyer authorization.
Every capture and annotation engagement runs under a documented consent and IP framework. Contributors sign individual contributor agreements (ICFs) that establish scope, compensation, and the IP assignment chain to the buyer. Capture protocols include consent steps for any identifiable subjects, geographies, or premises that require them. Documentation is auditable and retained for the life of the dataset.
Standard MSA and DPA terms are ready for most engagements; bespoke terms are accommodated where buyers have specific procurement requirements. Engagement starts when the contract is signed, not before.
Operational backbone runs through India-based delivery infrastructure: a registered legal entity, an in-house operations and quality team, and the annotator and reviewer pools that support most active programs. US-based account and client-services functions sit alongside delivery operations.
All annotators and reviewers admitted to active programs are background-checked through our standard onboarding flow. Buyers who require additional checks for sensitive engagements (e.g., domain-specific regulatory clearance) can specify the requirement in scoping; the additional vetting cost is part of the project budget.
Most engagements start with a short conversation about scope, geographies, languages, and timeline. We'll respond with a written proposal within a few business days.