An Alexa skill to answer questions about municipal services in Boston. Currently supports providing an address and asking for trash/recycling pick up days.
In order to create an Alexa skill we need to configure two main components:
-
The Alexa skill itself.
This is done on the Amazon developers page (https://developer.amazon.com). Here you need to configure an Alexa skill, allowing Alexa to understand and react to user voice commands.
-
An Amazon Web Service (AWS) lambda function to run our application logic.
This is done on the AWS console (https://console.aws.amazon.com/lambda)
Finally, we need to connect the Alexa skill to run our lambda function when it is activated.
The following instructions will walk you through creating the Boston Data Alexa skill, a new lambda function containing the Boston Data application, and then connecting the two together.
NOTE: The UIs for some of the consoles used below change frequently, however the general workflow is the same.
Clone this repo and navigate to the voiceapp311/mycity/mycity/deploy_tools directory and run the following command using Python 3:
python deploy_tools.py -p
This will generate the lambda_function.zip archive (two levels up in the directory), which you will need later.
- Go to the Amazon developers page (https://developer.amazon.com) and log in.
- Create a new Alexa Skill and name it Boston Data.
- In the Choose a model section, choose Custom and choose Create skill. Leave everything else as is, click Save at the bottom, then Next.
- In the Interaction Model section of the skill:
- In Invocation name section, put boston data.
- In the JSON editor input box, paste the contents of en_US.json.
- Leave the Slot Types section blank.
- Click Save model
- Click 3.Build Model on the right or Build Model on the top.
- Leave this tab open as you will enter information from your Lambda function into the Configuration section later.
-
Open a new tab and log in to AWS (https://aws.amazon.com)
-
Open a new tab and navigate to the AWS console again. Search for IAM:
- Select Roles.
- Select Create role.
- Under AWS service select Lambda.
- Under Permissions search for basic and you should see AWSLambdaBasicExecutionRole. Select it.
- Name your role lambda_basic_execution (it can be anything, but this is the name we use).
-
Search for Lambda and select Create function (and Author from scratch).
-
Make sure you're in US East (N. Virginia) region (top right) - this is the region for which Alexa Skill Kit is available.
-
Under Author from scratch info:
- give your Lambda the name BostonData
- For the runtime select Python 3.6
- For the role, select Choose existing role and select lambda_basic_execution (the role we created above).
-
Under Triggers, click Alexa Skills Kit from the menu on the left and it will appear in the empty dashed rectangle.
- You will also need to enter the Skill ID (also called Application ID). This can be found on the developer.amazon.com console for your skill. (go to your skill list by clicking Your Skills, you can see it under the skill name).
-
Click the uppermost box in the tree named BostonData (next to the orange icon) to open and modify the Function code section:
- Code entry type: Upload a .zip file pulldown.
- Click on the "upload" button, and browse to and select the lambda_function.zip archive created earlier.
- Runtime: Python 3.6
- Handler: lambda_function.lambda_handler
- Code entry type: Upload a .zip file pulldown.
-
Save.
-
In the upper right you'll see a ARN. Copy this and go back to the tab you have open from Part 1.
-
Set up a Google API key environment key.
Some of the intents require access to Google distance matrix which requires an access key to the Google API. In order to run these skills you will need your own access key which can be created by going to the Google Distance Matrix developer site and clicking the "Get Started" button. The distance matrix API is now grouped under the "Routes" API product.
Once you have a key, go to your AWS lambda configuration page. Find the environment variable section. Create a new environment variable with the key
GOOGLE_MAPS_API_KEY
and the value should be your personal key.
- Click 4. Endpoint on the left.
- In Service Endpoint Type, select AWS Lambda ARN and paste the ARN from part 2 into the default region field.
- Leave everything else as is.
Once you've completed Part 3, go to the Test tab. Here you can run sample voice queries using the Service Simulator. See usage below for some examples to test out.
Supports three custom intents right now.
- My address is [your Boston address]: Sets your address in the session.
- What's my address?: Tells you the current address in the session.
- When is trash/garbage/recycling day?: Tells you the trash/recycling days for the address in the session.
This section contains general information about how Alexa skills are executed and the services our intents interact with.
This is basically a restatement of the guide at https://moduscreate.com/build-an-alexa-skill-with-python-and-aws-lambda/ with some additional clarification. It traces the path of execution as a user interacts with our skill.
-
User issues voice command to Echo by saying, "Alexa" followed by a skill name and an intent. The intent may have parameters.
In this case, something like: "Alexa, ask Boston Data when is trash day?"
skill name : Boston Data intent : find trash days parameters : 1 Main Street apartment 2
We will give this intent a name, TrashDayIntent. You can see this in the intent schema.
-
Echo sends the request to the Alexa Service Platform.
This handles the speech recognition and translates the above voice command to a JSON document containing the intent and any parameters.
This JSON is sent to the skill (Boston Data in this example).
intent : trashday parameter : "1 Main Street apartment 2"
-
The skill receives the JSON.
We're implementing the skill as an AWS Lambda, so the JSON will be sent to the Lambda function at the ARN associated with the skill name.
-
The Lambda contains custom code that parses the JSON to identify the intent and corresponding arguments (in this example, the address).
The code then gathers data for the response. In this case that means a call to data.boston.gov to get the string of trash days associated with the provided address. Alternately this might mean accessing a database or session information.
This response data is serialized in a JSON response, which is returned to the Alexa Service Platform. It contains the response both as text for Alexa to say and as text/images for the smartphone app to display.
-
The Alexa Service Platform receives the response and conveys to the user using text-to-speech or the app display.
Because the python code in Boston Data's Lambda function uses external libraries, it must be uploaded as a .zip file.
To generate this .zip file, we must install all of the required Python packages in the directory that contains our code. For this project, that directory is BostonData/lambda_function. Amazon provides instructions on how to do so: https://docs.aws.amazon.com/lambda/latest/dg/lambda-python-how-to-create-deployment-package.html
Once all the requisite libraries are installed, compress the contents of the directory. The instructions note:
Important: Zip the directory content, not the directory. The contents of the Zip file are available as the current working directory of the Lambda function.
Recall that in Part 2 of the installation instructions we set the Handler to lambda_function.lambda_handler. This is specifying the function that is executed when a voice command is issued to the Alexa device. If we compress the containing directory instead of its contents, this code is not available.
This is the new (third party) source for Boston's trash day address information.
Boston's trash day schedule here: https://www.boston.gov/trash-day-schedule
labelled 'Powered by ReCollect'.
trash_intent.py accesses the information at recollect.net
jmartini noted on 02/13/2018 in issue #39: "The Boston digital team stated that the trash day info at data.boston.gov is often out of date. They are currently using a 3rd party called ReCollect to manage the newest trash day info."
This is the new portal for Boston's open data efforts. This site uses a tool called CKAN, which describes itself as follows:
CKAN is a tool for making open data websites. (Think of a content management system like WordPress - but for data, instead of pages and blog posts.) It helps you manage and publish collections of data. It is used by national and local governments, research institutions, and other organizations who collect a lot of data.
CKAN sites are organized by datasets, which can be made available in multiple formats. AnalyzeBoston contains 131 datasets as of this writing. Searching the datasets for "trash" returns 4 results, including one we can use for this project: trash schedules by address. The API for that dataset can be found by following the link to the dataset and then selecting preview. On the preview page there is a DATA API link.
Here you can find a link labelled odata. odata is a REST protocol for open data. CKAN provides an odata endpoint for the trash schedules by address dataset.
This is the Open311 portal for the city of Boston. It requires an API key and solely provides 311 functionality (checking the status of 311 issues or filing new ones). At some point we'll probably want to incorporate this type of functionality, but for the simple city service information retrieval we're concentrating on right now, this API is not useful.
DEPRECATED
This is the former open data portal of the city of Boston. This API implemented the Socrata Open Data API. Socrata is a company that provides cloud-based data visualization and analysis tools for opening government data. This portal is no longer being updated.
From Amazon's documentation:
Amazon Resource Names (ARNs) uniquely identify AWS resources. We require an ARN when you need to specify a resource unambiguously across all of AWS, such as in IAM policies, Amazon Relational Database Service (Amazon RDS) tags, and API calls. https://docs.aws.amazon.com/general/latest/gr/aws-arns-and-namespaces.html#arn-syntax-lambda
Our skill will be stored in an AWS Lambda function, which we can identify by its ARN.
Slots are used for intents that require parameters. Each slot must have:
- a name (a string describing the slot)
- a type (this can be a type preconfigured by Amazon or a custom type)
The preconfigured slot type for a street address is described here.
Information on defining a custom slot type is available here.
Alexa needs a list of phrases that correspond to each of our skill's intents.
We provide this in the settings for our skill in the developer console (see Part 1 of installation).
The format for the list of sample utterances is
[intent] [phrase]
The phrase may contain a reference to a slot if there is one associated with the intent it invokes. The format for this is
{slot_name}
Example of a sample utterance:
SetAddressIntent my address is {Address}
Our skill will invoke an Amazon Lambda function. This is where the code that produces a response to the Alexa voice command resides.
There are several language options for this code, including Javascript (Node.js), Java, C#, and Python(2.7 or 3.6).
Selecting Python we are provided the following template:
def lambda_handler(event, context):
# TODO implement
return 'Hello from Lambda'
The event argument is the JSON received from the Alexa platform. It contains the intent and slot information from the voice command.
Structure of the event object:
session
sessionId: [session id],
application
applicationId: [application id]
attributes: {},
user
userId: [user id]
new: true
request:
type: [request type, e.g., IntentRequest]
requestId: [request id]
timestamp: [timestamp]
intent
name: [name of the invoked intent]
slots:
[slot name]
name: [name of the slot]
value: [value of the slot]
locale: "en-US"
version: "1.0"
The elements of this event object are discussed in detail at: https://developer.amazon.com/docs/custom-skills/request-and-response-json-reference.html.