Zendesk AI Ticket Classifier

This is a project to classify customer requests based on the customer request text using AI - NLP. I have used AWS Comprehend, AWS Step Functions and Lambda Functions. In addition to these components, Zendesk supports natively connection to the AWS Event Bridge, this means that all ticket events in Zendesk are sent to AWS to process and update the ticket. You can consider and Event Based Architecture for your AI project as well.

1. Lambda Config

Each project lambda folder should be a different lambda function on your AWS project config. You will need to add the following environment variables for each lambda function as described bellow:

zd-sentiment-analysys-set-priority

• ZendeskAPIToken: You will need to create a Zendesk API Token and add here
• ZendeskEmail: Zendesk username used to create the API Token
• ZendeskSubdomain: without .zendesk.com

zd-sentiment-analysys-set-tags

• ZendeskAPIToken: You will need to create a Zendesk API Token and add here
• ZendeskEmail: Zendesk username used to create the API Token
• ZendeskSubdomain: without .zendesk.com

zd-sentiment-analysys-set-tags

• ZendeskAPIToken: You will need to create a Zendesk API Token and add here
• ZendeskEmail: Zendesk username used to create the API Token
• ZendeskSubdomain: without .zendesk.com
• ZendeskSLAHighWait: time in seconds to scalate the ticket
• ZendeskSLANormalWait: time in seconds to scalate the ticket
• ZendeskSLAUrgentWait: time in seconds to scalate the ticket

zd-sentiment-analysis-get-topic

• ClassifierEndpoint: Endpoint from the Comprehend custom classifier

2. Step Functions

Create an step function - state machine - by importing the JSON file aws-sentiment-analysis/step-function/ZendeskSentimentAnalysis.asl.json

3. AWS Comprehend - Custom Endpoint

In order to create your custom endpoint you will need to extract your ticket data, previously classified for your agents, in order to train the model to learn your categories in relation to the text from the ticket. You can use just the Subject and Description from your ticket or all the conversation.

4. Zendesk App

• zd-sentiment-analysis-app: You will need to compress the app files in order to upload on your environment. Feel free to make changes or improve this app. This app is just used for demos and examples, please do not use in production environment.

Released App: https://github.com/fabiolgc/aws-sentiment-analysis/releases/tag/zendesk-app-v1.0

To use in production please follow the Zendesk App Framework recommendations to avoid vulnerabilities and future security problems. Be careful please!!!! :)