Chatbot

• chatbot-client : https://stormy-dusk-60882.herokuapp.com/ (Primary - GUI)
• chatbot-server : https://desolate-wildwood-05365.herokuapp.com/
• chatbot-rasa-nlu : https://fast-escarpment-15453.herokuapp.com/

About

A financial assistant that serves basic banking functionalities to the customers.

To build locally

NOTE: You must have docker installed to build locally

To run the application locally, clone the repo and build using:

docker-compose build

Once build is complete, you can launch it by giving:

docker-compose up

You can then launch the app in your browser with the URL: http://localhost:3000

Tools used

• Database - MongoDB
• Front End - ReactJS
• Backend - NodeJs
• NLP - Rasa

Each of the components are containerized as individual docker images.

Architecture

The architecture diagram is as follows:

1. The customer sends a message from the client (browser)
2. The server (NodeJS) sends the message to the NLU Server.
3. NLU Server responds back to the server either the response message or an action.
4. (Optional) If needed, the server fetches the required data from the DB.
5. (Optional) DB returns the requested data back.
6. The server then sends the final data as response back to the customer.

Security

The client (ReactJS) and the server (NodeJS) communication is secured by using JWT.

When a new customer starts up the conversation, a new token is generated. The generated token is then stored in local storage for further communication.

When a user resumes the chat-session before the token expiry, the token from the local storage will be validated (using validation API shown below) and then used for communication.

If the user's token is expired, it will be re-generated and used for communication.

Upon successful validation of token, the chat history for the current user will be fetched and displayed in GUI.

Token expiration policy

Current token expiry is set to 2 hours. Once the token is expired, then a new token has to be generated again using the API shown below.

Auth APIs

The NodeJs server exposes the following two endpoints for token generation and validation respectively:

/v1/jwt_auth/new_session

/v1/jwt_auth/verify_session

JWT Access Token Generation

Used the online generator https://www.csfieldguide.org.nz/en/interactives/rsa-key-generator/ to generate private & public key.

Metrics APIs

The following APIs are used to obtain the metrics data

User Engagement

It shows the people who have engaged with the bot.

/v1/dashboard/user_engagement

Sample response:

{
  "engaged_users": [
      "Wayne",
      "Kent"
  ],
  "engagedUsersCount": 2,
  "total_users_visited": 5,
  "engagementRate": "40.00%"
}

Drop offs

It shows the people who got dropped off (closed the browser) or disconnected before the conversation is complete.

/v1/dashboard/drop_offs

Sample response:

{
  "dropped_users": [
      "Peter Parker",
      "Tony Stark"
  ],
  "droppedUsersCount": 2,
  "total_users_visited": 10,
  "dropOffRate": "20.00%"
}

Completion

It shows the people who completed a chat conversation without dropping off.

/v1/dashboard/completed_users

Sample response:

{
  "completed_users": [
      "Barristan Selmy",
      "Arthur Dayne",
      "Sandor Clegane"
  ],
  "completedUsersCount": 3,
  "total_users_visited": 18,
  "completionRate": "16.67%"
}

Sample Conversation

Bot conversation visualization

The bot conversation graph is as follows.

Accessing Mongodb & Schema

docker exec -it mongo-chatbot bash