Tutorial: schema-validated extraction with a retry loop

Before you start — about 45 minutes · Zod · JSON Schema · a JSON-mode model

Ship this in an afternoon. Zod does the work — the discipline is in the limits you set around it.

Step 1 — Decide what "working" means

Before touching code, write down three numbers: worst acceptable latency, cost per request, and what should happen when the upstream is unavailable. Everything after this is implementation.

Step 2 — Wire Zod

The whole flow is one path. No special branches for "large" inputs — express those as different limits, not different code.

const Invoice = z.object({
  vendor: z.string(),
  total: z.number().nonnegative(),
  dueDate: z.string().regex(/^\d{4}-\d{2}-\d{2}$/),
  lines: z.array(z.object({ sku: z.string(), qty: z.number().int().positive() })),
});

// validate first, persist second — never the other way round
const parsed = Invoice.safeParse(await extract(text));
if (!parsed.success) return retryWithFeedback(parsed.error);

Why this works

The limit is enforced before the expensive call, not after it. Reserving budget up front means a burst degrades the caller that caused it instead of everyone downstream.

Step 3 — Guard the failure path

asking for JSON in the prompt is not the same as constrained decoding

Write that guard before you write the feature. It costs an hour and saves a postmortem.

Step 4 — Verify it

npx tsx scripts/eval-extraction.ts --n 200   # prints violation rate

• Lower the limit and confirm the command fails loudly
• Kill the upstream and confirm you get a typed error, not a timeout
• Check the metric you emit per outcome: ok, throttled, upstream_error

Tuning knobs

Knob	Default	What we run	Why
concurrency	8	24	Most of the wait is network, not compute
retry budget	infinite	2, with jitter	Unbounded retries turn an outage into an avalanche
timeout	30s	8s, inherited from the caller	Nobody waits 30s for a chat reply

Copy it, change the limits, put the verification in CI, move on.