Project Atlas · Chase Card Data & Analytics
Stand up a governed analytics environment for a new card partner - in days, not a quarter.
Share program insight with a co-brand partner without ever leaking cardholder PII - by onboarding them as a first-class data-product domain with row/column entitlements, ML, and GenAI on curated data.
Switch role + purpose and watch rows scope and PII mask, live on a real database. This is the proof the whole design is built around.
The problem this artifact is designed against
Design a governed analytics environment for a new strategic card partner.
On a data lake / mesh, with hard row- and column-level entitlements and ML pipelines.
With generative AI built on top of curated data products - never the raw lakes.
Explore the surfaces
Lakehouse → medallion → data products → governance/entitlements → ML → GenAI. The reference architecture, with the two enterprise stacks.
The worked 'new strategic card partner' design: data-sharing model, 7 curated domains, onboarding runbook, partner reporting.
Curated products with field-level contracts, lineage, and entitlement-scoped previews, plus three deep-dive briefs.
Real Postgres RLS: switch role + purpose, watch rows scope and PII mask, live. The runnable governance proof.
Genie-style governed NL-to-SQL, portfolio dashboards, and the GenAI-on-curated-products pattern.
Live ML model cards - intended use, disaggregated performance, SHAP drivers, fair-lending posture, and SR 11-7 lifecycle stage. Sits under Platform / ML governance.
Vision & roadmap, flagship PRD, sprint board, and the SR 11-7 → 2026 model-risk register.
Product principles
Controls are designed into the data product - row scope, purpose, masking, audit - not bolted on after.
One source of truth with lineage by construction. Consumers read governed products, not extracts.
Every product ships a field-level contract: types, SLA, quality rules, allowed roles, purpose.
Who AND why. Fraud detail unlocks only under a Fraud Prevention purpose - even for fraud-capable roles.
What's real here, and what isn't
tap any item for the detailThe scenario
Atlas onboards a new strategic co-brand partner - illustrated here as a national retail/airline-class program, codename “Partner Aurora.” Aurora is a fictional stand-in used to make the design concrete; all cardmember data is synthetic and tokenized. The market context that frames it - Chase as the #1 US issuer and largest co-brand issuer - is real and cited on the Partner Environment surface.