This repository contains some of the product features I have learned over the last couple of weeks. It was inspired by an OfferZen make day. The scope of the solution was to create a chat bot application that would enable a person to talk to the Root Insurance Api and get a quotation on various offerings on Root's insurance platform. Since I have not worked with Google's DialogFlow yet, we decided (we were broken up in groups of 2 to 3 persons) to use DialogFlow as the chat platform. Secondly I wanted to play with Docker and see if I can put down a .net Core Web Api on Linux and use this as the communication link between the chat bot and the Root Api. Since I needed to save the state of the chat bot interactions, I also decided to play around with RethinkDB hosted in Docker.

I also made use of ngrok and will definitely use this awesome tool again in future.

As a side note, the .net Core applications was built in Visual Studio 2017, and I am running on a Windows 10 platform with docker installed.

This project is by no means complete yet and I think this might just be the starting point of it, but I think I did manage to get all the enviroments and Api's working without to much problems.

The first project is a chat bot created in Google DialogFlow. To get started with this, you need to create an account on Google DialogFlow. Once this is done you can create a new agent (don't use the V2 of the Api if you want to use the other projects). Now you can click on the gear icon next to your Agent name on the left hand side and click on the "Export and Import" button. Download the "DialogFlow Bot" chat bot folder, and Zip the contents of this folder. Now you can import it into DialogFlow just to see how I did some things. After this is done, you can go to the fulfillment tab and enter your Web Api Address in there. Please check the section on ngrok to expose your site on https as this is a requirement

The UI of DialogFlow and the documentation provided does not really match up that well. I also found that you should use one primary Web Browser Window to do the editing and saves of the project. I have not proved it, but it seems that the model is cached in the browser Window/Tab and if you click save on another older version of the project, it will overwrite some of the changes to older versions of the project.

Use the entities feature especially for yes/no questions. In this project there is a boolean entity with most of the yes and no answers a person can give. The entity value that gets passed back to the Api is then translated automatically into true or false and this is quite usefull for Json.

When asking for numbers, the DialogFlow seems to accept 15K or 15 thousand as a number, but I couldn't quite get it to work with million. I am not sure if I did something wrong but it did accept 15M as 15 million.

Then, try not to use follow up intents. If you start changing the contexts, the parent/child relationships keeps disappearing. It should in theory still work but if you are a bit OCD like me it might just ruin your day. Rather structure your project and intents to use Contexts. This project has a main context named "root_context" that is available throughout the project and every answer the user gives will be recorded in the "root_context". The way this works is simply as follows. On intents, you have input and output contexts. The intent(s) will become active as soon as all of its input contexts are available. Next, once an intent is satisfied with the response given by the user it will output the contexts specified. You are also able to clear output contexts by clicking on the number of the context and simply changing the value to 0.

The easiest way I got this working is by asking the question in the previous intent and listening for an answer in the next intent. This can get a bit confusing so I suggest that you create your intents with a prefix to group them accordingly. Also remember to give exact options on what the user is able to ask for, and try to keep the training as simple as possible. If there are two possible values with the word "insurance" in it, do not include that in the training engine as both options could pick this up and you might not get the flow of the dialog correct.

As a simple example, you would typically start with this.

• In the Welcome intent, you would typically greet the user and give a choice of options available. In the context out of this intent you would typically create your main context and the follow up context e.g. "root_context" and "option_context"
• Here there would be multiple intents (one for each option) that gets activated by the input contexts "root_context" and "option_context". On each of these intents the output context would be "root_context" and remove the "option_context" by setting the value to 0
  • Once any one of these option intents is satisfied, You would ask your first question regarding the question. You would then also add a new output context specific to the option e.g. "option_1_context" or "option_2_context"
• From here on out, just rinse and repeat.

The RootInsurance.SDK project is a wrapper around the Root's Insurance Web Api. This project was developed in .net Core 2.0 and still needs some work as not all of the calls has been implemented / tested yet. The Basics for the example that works with the chat bot is implemented. Here is the list of features currently supported.

NOTE THAT OF THE CURRENCY VALUES ARE PASSED IN AS CENTS AND NOT RANDS

• Get Gadget Models (Returns a list of gadgets that can be insured)
• Get life cover quotation
• Get funeral cover quotation
• Get gadget quotation
• Create a Policy Holder
• Get Policy Holders
• Get Policy holder by Unique Id
• Update a policy holder

There is quite a list of outstanding features that needs to be implemented and tested before the Api Wrapper is complete.

To use this library it is as simple as this:

// Remember to set the sandbox mode to true, the default value is false and the wrapper will then call the live site
var rootApiConfig = new RootInsurance.SDK.Configuration.Config()
{
    SandboxMode = true,
    RootApiKey = "Your Root Api Key Here"
};
var rootApi = new RootInsurance.SDK.ApiManager(rootApiConfig);
var gadgetModels = rootApi.Quote.Gadget.ListQuoteGadgetModels().Result;

To spin up a new RethinkDB instance on Docker, the easiest way I found was by using the Kitematic tool. On newer versions of Docker this can be installed directly from Docker. If you are running an older version of Docker, you download Kitematic you can just go ahead and copy the folder into the Installation Folder of Docker. The default path should be "Drive:\Program Files\Docker".

When docker is started, you can right click on the docker icon and then click on Kitematic to open the application. Once this opens, just search the templates for RethinkDB and this will download and create a RethinkDB instance in Docker. Once you start this instance, it will provide you with the Rethink DB management UI on one of the ports as well as the RethinkDB Api on another port.

You can check out this YouTube video on how to Get Started

What can I say about this awesome tool. In short, it exposes your localhost port to a public http/https port. This makes it a really easy way to debug your code. The reason I needed this is so that DialogFlow can post to my localhost projects without me having to set up a server with a static IP or anything. Just start the application, and point the DialogFlow to the exposed port on ngrok. This can be done by going to the Fulfillment tab, Enable the web hook and pasting the https address into the webhook address. Here is the documentation if you need some more information on options you can pass in but in all, I only used one of the options.

Run the application as an Administrator or at least from a command line started in administrator mode. If you are running the project in Docker, it is as simple as executing the following line

ngrok.exe http PORT

If you are running the project in IIS, you need to add the host header option when starting ngrok

ngrok.exe http -host-header="localhost:PORT" PORT