Your question is Secure Proxy With PII Redaction. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
Walk through the architecture of a secure proxy layer designed to sanitize inputs and redact PII before requests hit external LLM endpoints.
Explain the request and response data flow, including authentication, schema validation, PII detection and replacement, prompt safety checks, policy enforcement, audit logging, and routing to the Anthropic Claude API. Describe how you would handle false positives, detector failures, retries, idempotency, secrets, data retention, and observability without exposing sensitive content.