Building AI-Powered Features with the Claude API: A Practical Guide

Getting Started: API Key & SDK Setup

# Node.js
npm install @anthropic-ai/sdk

# Python
pip install anthropic

import Anthropic from '@anthropic-ai/sdk';

const client = new Anthropic({
  apiKey: process.env.ANTHROPIC_API_KEY,
});

Your First API Call

const message = await client.messages.create({
  model: "claude-sonnet-4-20250514",
  max_tokens: 1024,
  messages: [
    { role: "user", content: "Analyze this job listing for red flags: ..." }
  ],
});

console.log(message.content[0].text);

Prompt Engineering for Reliable, Structured Outputs

The key to production-grade AI features is predictable output. Always request JSON:

const response = await client.messages.create({
  model: "claude-sonnet-4-20250514",
  max_tokens: 2048,
  system: `You are a job listing analyzer. Always respond in valid JSON.
  Schema: { "risk_score": number (1-100), "flags": string[], "summary": string }`,
  messages: [{ role: "user", content: jobDescription }],
});

const analysis = JSON.parse(response.content[0].text);

Tips for reliable structured output:

1. Define the exact JSON schema in the system prompt
2. Use few-shot examples for complex outputs
3. Set temperature to 0.2-0.3 for analytical tasks
4. Validate the output with a schema validator (zod, ajv)

Building a Scoring Engine (Jobisque Example)

In Jobisque, Claude scores job listings across 5 dimensions. The trick is using weighted scoring:

const weights = {
  requirements_realism: 0.25,
  compensation_transparency: 0.20,
  company_legitimacy: 0.25,
  role_clarity: 0.15,
  red_flag_language: 0.15,
};

const overallScore = Object.entries(scores).reduce(
  (total, [key, value]) => total + value * weights[key], 0
);

Cost Management

| Model | Input (1M tokens) | Output (1M tokens) | |-------|-------------------|---------------------| | Claude Haiku 4.5 | $0.80 | $4.00 | | Claude Sonnet 4.6 | $3.00 | $15.00 | | Claude Opus 4.6 | $15.00 | $75.00 |

Cost optimization strategies:

1. Use Haiku for simple tasks (classification, extraction)
2. Cache system prompts — Anthropic's prompt caching reduces costs by up to 90% for repeated system prompts
3. Limit max_tokens to what you actually need
4. Batch requests when possible

Error Handling in Production

try {
  const response = await client.messages.create({ ... });
} catch (error) {
  if (error.status === 429) {
    // Rate limited — implement exponential backoff
    await sleep(Math.pow(2, retryCount) * 1000);
  } else if (error.status === 529) {
    // API overloaded — fallback to cached response or queue
  }
}

---

Want AI integrated into your product? Book a call and let's scope it.