Crowdin Integration

Prerequisites

Install the Custom MT Engines app from the Crowdin Store
Make sure this NMT platform is accessible from the internet (use ngrok or deploy to a server)
Have at least one completed training run for the language pair you want to use

Available Language Pairs

Project	Language Pair	Status
Gyant ES	`en` → `es`	No trained model

API Configuration

API Endpoint

Replace YOUR_SERVER with your server's public IP or ngrok URL

API Key

Keep this secret. It authenticates Crowdin to your NMT engine.

Crowdin Configuration

After installing the Custom MT Engines app in Crowdin, go to Machine Translation in your project settings, click Custom MT, and paste the following code snippets.

Request Code

// NMT Platform — Request Code for Crowdin Custom MT Engines
// Replace YOUR_SERVER with your server address (e.g., https://abc123.ngrok.io)

const NMT_BASE_URL = 'http://YOUR_SERVER:8000';
const NMT_API_KEY  = 'U8uZIEGWF7pV2kY-_3AWdukyy6REi5KrtRAXqGyY38s';

config = {
    url: `${NMT_BASE_URL}/api/crowdin/translate?source=${sourceLanguage}&target=${targetLanguage}&apiKey=${NMT_API_KEY}`,
    method: 'POST',
    headers: {
        'Content-Type': 'application/json'
    },
    data: {
        strings: strings.map((s) => typeof s === 'string' ? s : s.text)
    }
};

Response Code

// NMT Platform — Response Code for Crowdin Custom MT Engines

if (response && response.data && response.data.translations) {
    translations = response.data.translations;
} else if (response && response.error) {
    error = response.error.message || 'Translation failed';
} else {
    error = 'Unexpected response from NMT engine';
}

Test the API

Test the endpoint directly before configuring Crowdin:

cURL command

curl -X POST "http://YOUR_SERVER:8000/api/crowdin/translate?source=en&target=de&apiKey=U8uZIEGWF7pV2kY-_3AWdukyy6REi5KrtRAXqGyY38s" \
  -H "Content-Type: application/json" \
  -d '{
    "strings": [
      "Hello, how are you?",
      "The weather is nice today."
    ]
  }'

Expected response:

{
  "data": {
    "translations": [
      "Hallo, wie geht es Ihnen?",
      "Das Wetter ist heute schön."
    ]
  }
}

Troubleshooting

Make sure you have a completed training run for the language pair Crowdin is requesting. Crowdin sends language codes like en, de, fr — your project's source/target languages must match (prefix match, so en matches en-US).

Crowdin's servers need to reach your NMT platform over the internet. If running locally, use ngrok: ngrok http 8000 — then use the HTTPS URL ngrok gives you in the request code.

Translation quality depends on training data volume. With fewer than 10,000 sentence pairs, quality will be limited. For best results, use 100K+ parallel sentences. Check the Evaluation tab in your project to see BLEU/chrF scores.