AI CLASSIFIER

ai-classifier.ts

Classification with confidence thresholds, human fallback.

Stark

WHAT THIS PATTERN TEACHES

How to build AI classifiers with confidence-based routing: high confidence → auto-act, medium → soft-act with confirmation, low → human review queue. Includes fallback chains and drift detection.

WHEN TO USE THIS

Content moderation, ticket routing, sentiment analysis, intent detection — any classification task.

AT A GLANCE

async function classify(input: string) {
  const { label, confidence } = await model.classify(input);
  if (confidence > 0.9) return autoAct(label);
  if (confidence > 0.7) return softAct(label);
  return humanReview(input);
}

FRAMEWORK IMPLEMENTATIONS

TypeScript

interface Classification {
  label: string;
  confidence: number;
}

async function classify(input: string): Promise<Classification> {
  const result = await model.complete({
    messages: [{ role: 'user', content: input }],
    tools: [classifyTool],
  });
  return parseClassification(result);
}

function route(c: Classification) {
  if (c.confidence > 0.9) return autoAct(c.label);
  if (c.confidence > 0.7) return softAct(c.label);
  return humanReview(c);
}

← All Patterns