Design the Know-Your-Customer verification pipeline that decides whether a new applicant may open a financial account. An applicant submits identity documents and personal details; the pipeline orchestrates document checks, biometric/liveness matching, sanctions and watchlist screening, and risk scoring, then routes the case to auto-approve, auto-reject, or manual review. Because a wrong decision has legal and financial consequences, every verdict must be explainable and every input retained for audit.

Operating context. The pipeline calls several third-party verification providers that are slow, rate-limited, occasionally down, and priced per call. Applications arrive continuously with daily peaks. Documents contain sensitive PII that must be encrypted and access-controlled. Assume 500000 applications/day, provider calls that take seconds to minutes, and a regulatory duty to reproduce exactly why any past decision was made.

Out of scope. The applicant-facing onboarding UI, the internal implementation of third-party providers, ongoing transaction monitoring after approval, and the manual-review console beyond the queue it consumes. Assume separate teams own the app, the providers, and the review tool.

What to produce. A high-level architecture covering: the case orchestration across multiple async provider steps, handling slow/failing/rate-limited providers with retries and timeouts, the decisioning and routing to approve/reject/review, secure handling and retention of PII, idempotent re-runs and resumability of a case, an auditable record of every input and step, and scaling to peak load without exceeding provider quotas. Sketch the pipeline stages and dataflow; we will probe specifics in checkpoints.