Recipes | milliseconds.ai documentation

A pattern is a shape you apply to your own problem. A recipe is a finished job. Each page below names the capability, the exact labels or schema, the thresholds, and the code that ties them together.

Pick a recipe by the problem

Recipe	The job	Capabilities used
Support ticket triage	Route, prioritise, and flag an inbound ticket with three calls.	classify, rate, yes-no
Invoice extraction with verification	Pull a typed invoice record, then prove the critical fields.	extract, verify
Content moderation	Score harm, hard-flag the clear cases, send the rest to a person.	rate, yes-no
RAG passage filtering	Drop retrieved passages that do not answer the question.	yes-no
PII detection	Find personal data and its offsets, then redact.	entities
Taxonomy classification	Walk a deep label tree one level at a time.	classify

The recipes

Support ticket triage

Three independent calls return a queue, a priority score, and an urgency flag for one ticket.

Invoice extraction

Extract the invoice header fields, verify the number and total, build a review queue.

Content moderation

Rate harm on a described scale. Route the uncertain band to a human reviewer.

RAG passage filtering

Score every retrieved passage against the question. Keep, drop, or flag by probability.

PII detection

Describe each PII type, then redact with the start and end offsets the response returns.

Taxonomy classification

Classify level by level. Stop the descent when confidence drops below your bar.

By capability

Use this index when you already know the capability and want a worked example.

Capability	Recipes
Yes / no	Support triage, Content moderation, RAG passage filtering
Classify	Support triage, Taxonomy classification
Rate	Support triage, Content moderation
Extract	Invoice extraction
Entities	PII detection
Verify	Invoice extraction
Answer	No recipe yet. Choosing a capability covers when to use it.

By input type

Your input	Start here
A short message from a person	Support triage
User-generated posts or comments	Content moderation
A scanned or parsed document	Invoice extraction
Retrieved chunks from a vector store	RAG passage filtering
Free text that may hold personal data	PII detection
Text against a label tree of hundreds of labels	Taxonomy classification

Every recipe runs against https://api.milliseconds.ai as written. Every request needs an API key. See Authentication.

Before you copy one

Three things change per deployment. Read them once, then adapt each recipe.

Your label and statement wording drives accuracy. Writing good statements and labels gives the rules.
The thresholds in each recipe are starting points, not defaults. Set your own bands with thresholds and confidence routing.
A text holds at most 20,000 characters, and texts holds at most 32 items. Limits and rate limits lists every other field limit.