MatrAIx builds structured user representations that power simulation and evaluation infrastructure. Our internal corpus combines synthetic generation, evidence-aware extraction from human-authored public data, and consented self-reports, all mapped into one inspectable schema.
The corpus is not a single undifferentiated dataset. Generation method, source context, confidence, and evidence boundaries matter.
Personas are sampled from a typed attribute graph, with compatibility rules designed to reduce contradictory combinations and preserve diversity.
Public traces and consented survey responses are mapped to supported schema attributes. Unobserved fields should remain unknown rather than being invented.
Public biographical and professional context, mapped only where the source provides support.
BIOGRAPHICALPreference, product-use, and decision signals inferred from review histories with explicit evidence boundaries.
BEHAVIORALTechnical interests, tools, expertise, and developer-workflow signals derived from public contributions.
TECHNICALConsented, first-person responses provided directly by participants and mapped to the shared persona schema.
SELF-REPORTEDNo source makes every attribute observable. Sensitive or unsupported claims are excluded, and collection or release eligibility depends on consent, source terms, privacy review, and documentation.
We will not directly distribute the full 8.3B internal corpus. It includes large-scale synthetic outputs, human-evidence extraction artifacts, and consented self-reports that require source-aware governance, storage, and quality controls.
A curated subset is planned for Hugging Face with a dataset card, schema documentation, provenance summaries, validation notes, and stated limitations.
— dimensions across — categories, with — allowed values. Search, filter, expand, or sample a structured persona below.
Canonical schema ↗No dimensions match that filter.