Oh my god. It's full of code!

Making Amazon Echo/Alexa order me a pizza

UPDATE! I’ve started a github project for my server software. Still very alpha but you can check it out here: https://github.com/Kenji776/AlexaHomeHub

If you follow my blog, you might have caught my post yesterday about how I bought an Amazon Echo device, and have begun creating my own custom actions for it. I started small, simply making it call out to my web server by using If This Then That (IFTTT – a website that allows you to connect and integrate different services). I got it to connect to my door lock and unlock service I had written, and even got it to chromecast specific pre-setup videos to one of my TV’s using a command line tool. Feeling somewhat confident I decided it was time to take on something a little more in depth, but oh so worth it. I was going going to make Alexa order me a pizza.

If you are an Echo/Alexa user you might know that there is already support for ordering a pizza but only from Dominos. It uses some system of having a saved order, then tweeting a specific bit of info at the Dominos twitter account that is tied to your order which then places it. This has a few drawbacks. Primarily being that it orders you a Dominos pizza (sorry guys, in all fairness Dominos has gotten a lot better in recent years). Also it requires twitter integration, and as far as I know only supports one order that you have saved (I could be wrong). The using a saved order was a good idea as it streamlines and simplifies the ordering process quite nicely. I wanted to do something like this, but instead I wanted Sarpinos pizza, and I wanted to be able to pick from several different pre-created orders. Using my knowledge of browser automation that I picked up from my door lock/unlock project and my existing web server, I got to work.

First off, I had to figure out all the things that needed to happen. Off the bat I knew I’d be using their online ordering interface. They don’t have an API, so I knew I’d have to be automating browser interactions. Next I had to break down the process of ordering the pizza online step by step, all the HTML elements involved and how to interact with them. Then I would be able to automate those interactions using the Selenium library. So I went through the process like a normal person and created this list. At each step I inspected the HTML elements involved and recorded them so I could figure out how to identify them and interact with them later. I created an order and saved it as a favorite so next time I’d come back in I would be prompted if I’d like to order that again. From there I was able to create the following list of things I knew needed to get done.

Order Steps:

1) Invoke: https://order.gosarpinos.com/Login/

2) Wait For Load

3) On load populate credential fields:
	- <input class="text-box single-line valid" data-val="true" data-val-required="Email is required" id="Email" name="Email" type="email" value="" >
	- <input class="text-box single-line password valid" data-val="true" data-val-required="Password is required" id="Password" name="Password" type="password" value="" >

4) Click Login button
	- <input type="submit" value="Login" class="ui-button ui-widget ui-state-default ui-corner-all" role="button" aria-disabled="false">
5) Wait For Page Load

6) Find button with provided favourite id (428388)
	- <button class="wcFavoriteSelectFavoriteButton ui-button ui-widget ui-state-default ui-corner-all ui-button-text-only" data-favorite-id="428388" id="wiSelectFavorite428388" type="button" role="button" aria-disabled="false">

7) For For Delivery Popup to load


A fair amount of steps, but none of them super complicated. I knew I’d have to learn a bit more about Selenium as this interaction was definitely more complicated than the door lock code, and that one was already seemingly over complicated. Thankfully I did, and found out that in my previous attempt I had been mixing synchronous and async methods unknowingly hence leading to perceived complexity (I thought driver.wait() was a async method and you put everything that depended on it inside. Turns out it’s synchronous and once the condition inside is true the program continues. No wonder it was acting a little funny). I knew also since I was going to be passing in a fair amount of data (username, password, order id, credit card info, etc) that I should probably define an object which would have all the required properties, then just pass JSON into my web service that mirrored that object. This is what I came up with.



Obviously the sensitive values are blacked out, but you get the jist. My webserver is already primed to look for post requests that have a JSON payload. The ‘router’ code looks at the ‘action’ attribute to figure out what function to send the payload to. I created a new ‘pizza’ action and related function. Here is that function.

function orderPizza(orderObject,callback)
	var orderResult = new Object();
	//create instance of selenium web driver
	var driver = new webdriver.Builder().

	//request the login page with the locks page as the return url
	//find and click submit button

	console.log('Logged in as: ' + orderObject.username);

	//wait for order page to load
	driver.wait(function() {
		return driver.getTitle();
	console.log('Attempting To Choose Favorite Order With Id: ' + orderObject.orderId);
	//have to wait until the proper order button appears since it's in a dialog. If it isn't found after 5 seconds, fail. Otherwise click the corresponding order button
	//the favorite order button has an attribute 'type' of 'button' and a 'data-favorite-id' attribute with the id of that order
	driver.wait(function () {
		return driver.findElement(webdriver.By.css("div[aria-describedby='wiFavoriteListDialog']")).isDisplayed();
	}, 5000);
	driver.wait(function () {
		return driver.findElement(webdriver.By.css("button[data-favorite-id='"+orderObject.orderId+"']")).isDisplayed();
	}, 5000);
	//after the button above is clicked, that dialog closes and another one opens. This one asks the user to select delivery or pickup. We want delivery
	//the delivery button has an attribute with a 'data-type' of 'WBD' and an attribute 'role' of 'button'
	driver.wait(function () {
		return driver.findElement(webdriver.By.css("button[data-type='WBD'][role='button']")).isDisplayed();
	}, 5000);
	//ensure the checkout button is visible
	driver.wait(function () {
		return driver.findElement(webdriver.By.css("button[id='wiLayoutColumnGuestcheckCheckoutBottom'][role='button']")).isDisplayed();
	}, 5000);
	//hacky method to ensure that the modal dialog should now be gone and we can click the checkout button
	//then the browser will move to the order screen. Once it loads we have to populate the order field data

	//wait for payment page to load
	driver.wait(function() {
		return driver.getTitle();

	//wait until the pay by credit card button shows up.
	driver.wait(function () {
		return driver.findElement(webdriver.By.id("wiCheckoutPaymentCreditCard")).isDisplayed();
	}, 5000);

	//check the pay by credit card radio button
	//wait until the credit card number box shows up
	driver.wait(function () {
		return driver.findElement(webdriver.By.id("Payment_CCNumber")).isDisplayed();
	}, 5000);
	//populate the form fields

	//stupid jQuery ui selects are impossible to set with normal selenium since the original select is hidden. So use an execute script to set em.


	//wait for confirmation page to load.
	driver.wait(function() {
		return driver.getTitle();


	driver.wait(function() {
		return driver.getTitle();
		console.log('Ordering complete!');
		orderResult.success = true;
		orderResult.message = 'Order Placed Successfully';



Now if that code seems a little dense or confusing, don’t feel bad. It took me several hours of trial and error to figure it out, especially then it came to setting the select list values, and getting the script to wait while various elements where created and destroyed by the page. Selenium has this super fun behavior where if you ever try and reference an element that doesn’t exist, the whole script goes down in flames. In response to that I made my code very ‘defensive’ checking to make sure elements that are required frequently before attempting to interact with them.

With the script created and integrated into my web server ‘router’ I was ready to get IFTTT to invoke it. Once again it was as simple as creating a new recipe with Alexa as the If and the Maker make a web request feature as the do.

pizza 1pizza2

You can see that with the combination of specific phrases and the fact that you can have multiple saved orders, it would be easy to setup many different possibilities. My roommate is even going to create his own IFTTT account and link it to my Alexa. Then he can create his own orders, specify his own credit card information in the JSON payload, and order whenever he wants using the same device but have his own information. The next step I think is to encrypt the JSON payloads which contain the credit card info and then decrypt them when the arrive at my server. That way I’m not storing my CC info in plain text anywhere which right now is a bit of a concern. This was mostly just proof of concept stuff last night, but I was too excited not to share this as soon as I could, so some of the ‘polish’ features are missing but overall I think it’s a damn good start. Now if you’ll excuse me, I’m going to get myself a pizza.

Update: Adding encryption was pretty easy. First I got some encrypt and decrypt functions set up. Like this.

// Nodejs encryption with CTR
var crypto = require('crypto'),
    algorithm = 'aes-256-ctr',
    password = 'xxxxxxxxxxxxxxxxxxxxxxxxxx';

function encrypt(text){
  var cipher = crypto.createCipher(algorithm,password)
  var crypted = cipher.update(text,'utf8','hex')
  crypted += cipher.final('hex');
  return crypted;
function decrypt(text){
  var decipher = crypto.createDecipher(algorithm,password)
  var dec = decipher.update(text,'hex','utf8')
  dec += decipher.final('utf8');
  return dec;

Then update my incoming data object so that the encrypted data was in its own property so I could still tell what kind of request it was without having to decrypt the payload first (since my webserver supports other, unencrypted calls).

"data":"f613f8f479bad299bdfedf [rest of encrypted string omitted]",


Then just had to change my ‘router’ to decrypt the incoming data if encryption was detected.

else if(action == 'pizza')
	//pizza request contains encrypted info. Decrypt and send to function
	var pizzaRequestData = new Object();
		console.log('Encryped Payload Detected. Decrypting Containted Data');
		pizzaRequestData = JSON.parse(decrypt(parsedContent.data));
		console.log('Decryption complete');

		responseObject.message = 'Ordering Pizza!';
		//because order pizza is async the result data comes in a callback
			responseObject.pizzaRequest = data;
		console.log('Un-encrypted pizza order detected. Skipping');

After that I just had to use the encrypt function to generate an encrypted version of my pizza request data, update the IFTTT recipe with the new request and that’s it! Now my CC information is safely encrypted and I don’t really have to worry about it getting intercepted. Yay security.

One response

  1. Great job!

    March 30, 2016 at 9:03 am

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s