Microsoft Presidio Alternative for Node.js: 0.8ms In-Memory vs 40ms Docker

// [BEFORE] Microsoft Presidio (2 HTTP Calls, Docker, 45ms+ latency)
/*
const analyzerRes = await fetch('http://presidio-analyzer:5001/analyze', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ text: userInput, language: 'en' })
});
const analysisResults = await analyzerRes.json();

const anonymizerRes = await fetch('http://presidio-anonymizer:5002/anonymize', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ text: userInput, analyzer_results: analysisResults })
});
const { text: sanitizedPrompt } = await anonymizerRes.json();
*/

// [AFTER] PrivacyScrubber SDK (1 In-Memory Call, 0 Docker, <1ms latency)
import OpenAI from 'openai';
import { sanitize, restore } from '@privacyscrubber/sdk';

const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
const userInput = 'Patient Johnathan Miller (DOB: 1984-03-12, SSN: 042-88-9124) requested clinical summary.';

// 1. Sanitize in local RAM before LLM dispatch (<0.85ms)
const { scrubbedText, tokenMap } = sanitize(userInput, {
  profile: 'General'
});

// 2. Invoke OpenAI with sanitized prompt (zero PII leaves process):
const completion = await openai.chat.completions.create({
  model: 'gpt-4o',
  messages: [{ role: 'user', content: scrubbedText }]
});
const aiResponse = completion.choices[0].message.content || '';

// 3. Deterministically restore original values in local RAM:
const cleartextResponse = restore(aiResponse, tokenMap).restoredText;
console.log(cleartextResponse);