Ask a question, get statistically-grounded answers from synthetic personas anchored in public census/Eurostat data — every insight wrapped in a conformal confidence bound.
This is a public showcase. The source is proprietary — evaluation access on request.
CHORUS generates synthetic personas statistically anchored to public population data (Census, Eurostat, SOEP-public), runs surveys and what-if questions against them, and returns aggregate insights with conformal-bounded confidence rather than a single hand-wavy number. It is designed for research, product, and policy questions where you need a defensible read on a population without touching a single real person's data.
- Persona synthesis — demographically-anchored personas with causal backstories, value vectors, and provenance trails (every generation hashed + timestamped + source-linked).
- Hybrid PPI++ — prediction-powered inference (Angelopoulos--Bates) blends a few real labels with many model predictions for an unbiased, lower-variance estimate.
- Conformal-bounded insights — aggregate answers ship with finite-sample confidence intervals, not point guesses.
- Replication-aware — designed to reproduce published study effects as a calibration check.
Re-identification risk is structurally zero: personas are synthetic, anchored only to aggregated public statistics. No personal data enters the pipeline — GDPR-clean by construction.
Pre-customer, private beta. Not production-hardened. Built on the open trust-core conformal primitives.
Source and a guided demo are available under a commercial/evaluation license on request.
Ozan Küsmez — ozan.kuesmez@outlook.com
© 2026 Ozan Küsmez. CHORUS is proprietary software, all rights reserved. This showcase repository contains documentation only.