PII Detector

Detects whether a message carries a real person's personal data or a real secret (password, API key), so it can be redacted before it reaches an LLM. Code does what regexes are good at: finding email-, phone-, card- (Luhn-checked), SSN- and key-shaped strings, and masking them. Whether a match is really a person's data (a customer's email or support@, a real card or Stripe's test card) and whether there is PII no regex can see (a name with a home address and a diagnosis, an email spelled out in words) is judgment: one Jev Noul, with a threshold tuned by the sweep. Measured: accuracy, PII missed, clean text flagged, and the threshold sweep.