mediumAI Engineering

How do you handle PII and sensitive data when calling LLM APIs?

Q: How do you handle PII and sensitive data when calling LLM APIs?

**TL;DR.** Verify the **provider contract** (no-training terms, zero/short retention, regional processing), then **minimize**: send only needed fields, **redact/pseudonymize** identifiers before the call, and **scrub logs and traces** — the surface teams forget. For the strictest data, keep inference **in-VPC**. raw record Rahul, +91-98... redact <text class='d-sub' x='198' y='

314 views

Understand the problem

Data minimization, redaction, retention terms and regional processing for LLM traffic.

piiprivacycompliancesecurity

Attempt it yourself

Sketch your approach before reading the solution — that's what interviews test.

Stuck? AI Nudge Available

Get a conceptual hint to guide your logic without spoiling the final implementation.

Study the solution

The solution is waiting

Give it an honest attempt first — then compare your thinking with the full walkthrough.

Read the code

Redact → call → restore

const { redacted, mapping } = piiRedact(ticketText);
// "Customer [NAME_1] ([EMAIL_1]) reports billing issue on [CARD_1]..."

const reply = await client.messages.create({
  model, max_tokens: 500,
  messages: [{ role: "user", content: draftReplyPrompt(redacted) }],
});

const restored = piiRestore(textOf(reply), mapping);   // [NAME_1] → actual name
await traces.write({ prompt: redacted, output: scrub(textOf(reply)) }); // never raw

Join the discussion

Discussion (0)

No responses yet. Be the first to share what you think.

How do you scope retrieval with metadata filtering and multi-tenancy?

What is hybrid search and why combine BM25 with vector search?