Skip to main content

Architecture

img alt

AI Studio is a multi-tenant system that supports the creation of three types of projects: NLU, KM, and Cognibot. The cognibot project lets you build a chatbot that can utilize natural language understanding and knowledge article search capabilities of NLU and KM projects respectively. The cognibot project supports connecting to AutomationEdge as a fulfillment engine for your Information Technology Process Automation (ITPA) or Robotic Process Automation (RPA) automation requests.

  1. AI Studio-UI: It is the administrative console that lets you create three types of projects mentioned above. It also has a dialog designer that lets you build and test chatbot dialogs or conversations without writing a single line of code. It talks to AI Studio-Engine using Representational State Transfer (REST).
  2. AI Studio-Engine: It is the backend of AI Studio. It uses a relational database as the data store.
  3. NLU Engine: It gives you the ability to search structured documents. While creating an NLU project in AI Studio, you have to choose the NLU engine. In this release, we only support LUIS as the underlying engine.
  4. KM Engine: It gives you the ability to search unstructured documents. While creating a KM project in AI Studio, you have to choose the KM engine. In this release, we only support QnAMaker as the underlying engine.
  5. Channel Connector Service: It manages communication between chatbot web service and different chat channels. It relieves the chatbot web service from having to manage communication between different chat channels like web chat, MS Teams, FB messenger, etc. which means, the chatbot web service code pretty much becomes channel agnostic.
  6. Chatbot Web Service: It manages chat discussions, conversation state, etc. It maintains a record of each chat conversation using a relational data store. Additionally, it interacts with AI Studio to talk with the underlying NLU and KM projects. Both the chatbot web service and the channel connection service can be deployed separately. The endpoints of both of these services must be configured in a Cognibot project for this reason. However, while setting up an on-premise installation of AI Studio, we would deploy both of them as a single service because doing so is convenient for development. This single combined service will be referred to as a cognibot web service in the sections that follow.
  7. AI Studio-DB: PostgresSQL database is used by both AI Studio-Engine and chatbot web service. Typically, you create two different databases for these.

On-premise Deployment​

Installation section covers on-premise deployment of all the components shown in the above architecture diagram. As we would be using LUIS and QnAMaker as NLU and KM engines respectively, you don't need to install them locally as they are available only in the cloud. You can connect to your own LUIS app or QnAMaker knowledge base by providing appropriate connection details. But you can also connect to our LUIS app and our QnAMaker knowledge base for getting things up and running quickly. You will have a complete cloud trial experience in the upcoming release of AI Studio.

Integration with AutomationEdge​

AI Studio Cognibot project uses AutomationEdge for fulfilling the requests originated by users interacting with the chatbot. We have created a hosted AutomationEdge instance with some predefined workflows, using which you can avoid setting up an AutomationEdge instance for the trial. This instance https://t4.automationedge.com running inside AWS is managed by us. Workflows that are called from the chatbot typically repack by making a REST call on the chatbot web service as shown below. img alt
Assuming you are running chatbot web service on port 3978 on the local machine, its endpoint http://localhost:3978 will not be accessible over the internet. In order to solve this issue and to integrate your chatbot web service with MS-Teams as will be covered in the section MS-Teams, you can use ngrok which lets you create tunnels to localhost. With ngrok, tunneled service is given a public end-point. If you run  ngrok http 3978 command on the local machine, you will see an end-point similar to  https://cdab-116-74-169-47.ngrok.io which will have to be configured in file .env of the chatbot web service with the key named CHATBOT_URL. One thing to note here is that every time you run  ngrok http 3978 command, you are going to get a different public end-point.
If you are running chatbot web service on a machine with public IP address and with incoming traffic to the port on which you are running chatbot web service allowed, you don't need ngrok for AutomationEdge workflow to communicate back to chatbot web service. But for integrating with MS Teams, you not only need a public end-point but also has to need a valid https certificate. ngrok gives you that as well.