Custom NLP

Quickly set up your custom NLP into your NLX workspace

Why use a custom NLP?

Off-the shelf NLP providers, such as Amazon Lex or Google Dialogflow, may not possess the functionality a business requires for handling conversations with its users. Large enterprises often resort to tailored solutions with a custom NLP model.

---

How does NLX handle my custom NLP?

Just as Amazon Lex or Google Dialogflow provide APIs to ensure their NLPs are compatible with conversational AI builders, NLX provides you with simple API specifications that allow your custom NLP to handle the same actions. These include the build and deployment of a bot as well as essential conversation runtime operations, such as disambiguating a user utterance.

Identical to using off-the-shelf NLPs, testing utterances through our automated test suite, architecting and building intent flows using the canvas builder, and tracking performance of conversations with analytics can all be done with a custom NLP.

After your custom NLP's API is made compatible and is integrated in NLX, you may select it as the engine of choice when deploying your bot.

---

Create compatible API

Your custom NLP's API interfaces with NLX to build your conversation flows and process user utterances.

Integration architecture

As a result, before you can link your custom NLP in NLX, an API endpoint must be configured using our specification:

Disambiguate unstructured text using NLP function

POST/disambiguate

Authorization

api_key

Body

application/json

buildIdstring

utterancestring

languageCodestring

en-US, es-ES etc.

stateobject

contextobject

context attributes available in NLX conversation

Response

200400500

Successful operation

Body

application/json

intentIdstring

slotsarray of object

sentimentenum

positivenegativeneutral

confidenceScorenumber

Request

JavaScriptCurlPython

const response = await fetch('/disambiguate', {
    method: 'POST',
    headers: {
      "Content-Type": "application/json"
    },
    body: JSON.stringify({}),
});
const data = await response.json();

Response

200

{
  "intentId": "text",
  "slots": [
    {
      "slotId": "text",
      "value": "text"
    }
  ],
  "sentiment": "positive",
  "confidenceScore": 0
}

Create a new NLP build

POST/builds

Authorization

api_key

Body

application/json

buildIdstring

signedUrlstring

Response

200400500

Successful operation

Body

application/json

buildIdstring

statusenum

builtdeployedcanceledpending_buildpending_deploymentfailed

errorMessagestring

Request

JavaScriptCurlPython

const response = await fetch('/builds', {
    method: 'POST',
    headers: {
      "Content-Type": "application/json"
    },
    body: JSON.stringify({}),
});
const data = await response.json();

Response

200

{
  "buildId": "text",
  "status": "built",
  "errorMessage": "text"
}

The signedUrl attribute in the request body refers to presigned S3 URLs for accessing the bot's build metadata. The URL expires in 5 minutes, and should suffice for downloading and caching the metadata within customNLP. customNLP may cache and use the build metadata for disambiguation requests. The artifact is a zipped file containing the following top-level files and directories:

1. intents - A directory with information about the bot's intents.

   • Includes a sub-directory for each languageCode.

   • Within each languageCode, there is one JSON file per intent containing metadata such as utterances, intentId, and slots, with each utterance translated to the respective languageCode.

2. slotTypes - A directory with information about the bot's slots.

   • Includes a sub-directory for each languageCode.

   • Within each languageCode, there is one JSON file per slot with metadata like values and synonyms translated to the relevant languageCode.

3. manifest.json - A JSON file with metadata about the build, including attributes like botId, buildId, supported languageCodes and createdTimestamp.

Retrieve status of a build

GET/builds/{buildId}

Authorization

api_key

Path parameters

buildId*string

Response

200400404500

successful operation

Body

application/json

buildIdstring

statusenum

builtdeployedcanceledpending_buildpending_deploymentfailed

errorMessagestring

Request

JavaScriptCurlPython

const response = await fetch('/builds/{buildId}', {
    method: 'GET',
    headers: {},
});
const data = await response.json();

Response

200

{
  "buildId": "text",
  "status": "built",
  "errorMessage": "text"
}

Update the deployment status of a build

PUT/builds/{buildId}

Authorization

api_key

Path parameters

buildId*string

Body

application/json

actionenum

deploycancel

Response

200400404500

successful operation

Body

application/json

buildIdstring

statusenum

builtdeployedcanceledpending_buildpending_deploymentfailed

errorMessagestring

Request

JavaScriptCurlPython

const response = await fetch('/builds/{buildId}', {
    method: 'PUT',
    headers: {
      "Content-Type": "application/json"
    },
    body: JSON.stringify({}),
});
const data = await response.json();

Response

200

{
  "buildId": "text",
  "status": "built",
  "errorMessage": "text"
}

Make sure to set up an API key for your API, as it will be required during the integration step. If you require private connectivity between NLX and your on-premises API, please contact your NLX Customer Success Manager.

Download full OpenAPI Specification:

5KB

nlx-custom-nlp-openapi.yaml

You can import and visualize the NLX OpenAPI specifications using Swagger Editor.

---

Integrate API

Once your API is ready for integration, navigate to the Integrations tab of your workspace's Settings:

• Click + Add integration

• Provide your integration a name

• Choose Custom NLP from the dropdown

• Enter both your API URL + API key

• Select Create integration

If desired, at any time you can edit or delete your custom NLP integration by expanding the integration name and selecting the edit or delete icons.

Your integration is now available for selection when choosing an engine under your bot's Deployment tab.