Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Actions with Google
Search
Bryan Herbst
February 04, 2017
Programming
1.5k
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Actions with Google
Learn how to create custom actions for the Google Assistant
Bryan Herbst
February 04, 2017
More Decks by Bryan Herbst
See All by Bryan Herbst
Semantics in Jetpack Compose
bherbst
0
13k
Kotlin Multiplatform
bherbst
0
990
Android App Architecture & Navigation
bherbst
0
2.1k
What's in the Android Toolbox
bherbst
0
570
Gradle Deep Dive
bherbst
2
2.2k
Android Transitions
bherbst
0
580
Databinding in Android
bherbst
0
480
Activities Vs. Fragments
bherbst
0
390
Android Transitions
bherbst
0
77
Other Decks in Programming
See All in Programming
DynamoDBの基礎を振り返りながらベクトル検索機能を理解する
musan
3
260
Can LLMs Replicate 4 Years of Compose Migration? Exploring the boundaries of automation with 279 XML files from a real product
makun
0
120
My Marp Sample
sinoue0108
0
140
バグを直したら useEffect が消えた
colorful12
3
790
ソフトウェアエンジニアにとっての生成AI - 特性を知って使い倒す / generative ai for software enginner
kishida
7
2.3k
初めての模倣学習とVLA
natsutan
0
400
in-process GraphQL のすすめ #ginzajs
izumin5210
4
1.5k
高専、大学編入、そして未踏へ〜プロダクト開発とキャリアの歩み - Technical College, University Transfer, and On to “Mitou” / My Journey in Product Development and Career
pkmiya
0
130
go-spidermonkeyでAIエージェントのCode Modeを実装する
syumai
3
1.5k
Discordを用いたラボオートメーション関連情報収集の自動化
noguhiro2002
0
480
新人はどこまで自力でやり、どこからAIに頼るべきか/エンジニア育成に向き合う_先輩たちの悩みと知見共有会
toppan_digital_dev
1
430
AIを上手に使っていこうとしたら越境せざるを得なくなった話 〜実践1年で見えた境界を越えなければならない理由と進め方〜 / Crossing borders with AI
tomoyakitaura
2
920
Featured
See All Featured
Building Better People: How to give real-time feedback that sticks.
wjessup
370
20k
Six Lessons from altMBA
skipperchong
29
4.5k
Building a A Zero-Code AI SEO Workflow
portentint
PRO
0
700
Bootstrapping a Software Product
garrettdimon
PRO
306
120k
エンジニアに許された特別な時間の終わり
watany
108
250k
Agile Leadership in an Agile Organization
kimpetersen
PRO
0
220
A better future with KSS
kneath
240
18k
CoffeeScript is Beautiful & I Never Want to Write Plain JavaScript Again
sstephenson
162
16k
More Than Pixels: Becoming A User Experience Designer
marktimemedia
3
510
Automating Front-end Workflow
addyosmani
1369
210k
Exploring the relationship between traditional SERPs and Gen AI search
raygrieselhuber
PRO
2
4.3k
How to Grow Your eCommerce with AI & Automation
katarinadahlin
PRO
1
260
Transcript
Actions with Google Bryan Herbst Android Engineer @ Target
The Google Assistant “Ok Google” Allo Google Home Android Auto
Android Wear
The Google Assistant Google Home
Design principles Voice interaction is different
What actions can I take?
What actions can I take?
What should I have for dinner? What are you in
the mood for? You can say Italian, American, Mexican, or Chinese. !
What should I have for dinner? What are you in
the mood for? You can say Italian, American, Mexican, or Chinese. ! Take turns.
“Guide me downtown.”
“Guide me downtown.” Downtown where? New York? Minneapolis?
“Guide me downtown.” Driving or walking? Starting from where?
Speech uses shortcuts Filling in the blanks requires context.
Go scriptless You are (not) a robot.
Freedom of speech Say “help” for more options
Freedom of speech Say “help” for more options
Anticipate errors Shortcuts can cause ambiguity
”Buy me a KitKat bar.”
”Buy me a KitKat bar.”
”Buy me a KitKat bar.”
”Buy me a KitKat bar.”
”Buy me a KitKat bar.” ”Okay, ordering a king sized
KitKat bar from Amazon. Is that correct?”
”Buy me a KitKat bar.” ”Okay, ordering a king sized
KitKat bar from Amazon. Is that correct?” “Yes.”
”Buy me a KitKat bar.” ”Okay, ordering a king sized
KitKat bar from Amazon. Is that correct?” “No, make it a regular sized KitKat bar.”
Anticipate errors Shortcuts can cause ambiguity
Be Flexible Shortcuts can cause ambiguity
Types of actions Conversation and direct
Conversation Direct Trigger “Talk to SERVICE about TOPIC” Responsibilities Availability
Conversation Direct Trigger “Talk to SERVICE about TOPIC” Responsibilities You
handle interaction and fulfillment Availability
Conversation Direct Trigger “Talk to SERVICE about TOPIC” Responsibilities You
handle interaction and fulfillment Availability Now
Conversation Direct Trigger “Talk to SERVICE about TOPIC” “Do this
action” Responsibilities You handle interaction and fulfillment Availability Now
Conversation Direct Trigger “Talk to SERVICE about TOPIC” “Do this
action” Responsibilities You handle interaction and fulfillment Google handles interaction, you handle fulfillment. Availability Now
Conversation Direct Trigger “Talk to SERVICE about TOPIC” “Do this
action” Responsibilities You handle interaction and fulfillment Google handles interaction, you handle fulfillment. Availability Now Soon
Conversation Actions " Less ambiguity # Limited trigger
Making Conversation Actions
Conversation Actions Format $ Your Action Input
Conversation Actions Format $ Your Action Input JSON Request
Conversation Actions Format $ Your Action Input JSON Request JSON
Response
Conversation Actions Format $ Your Action Input Action Response JSON
Request JSON Response
Conversation Actions Format Your Action JSON Request JSON Response Conversation
API
Invocation Dialogs Fulfillment “Ask AwesomeFlightApp to book me a flight.”
Invocation Dialogs Fulfillment “Ok, where would you like to go?”
“Paris.”
Invocation Dialogs Fulfillment “Okay, booking a flight to Paris!”
Invocation - name Unique
Invocation - name Unique | Easy to say
Name restrictions Two words (with exceptions)
Name restrictions Two words (with exceptions) | Avoid reserved words
Trigger Phrases • Let me talk to… • I want
to talk to... • Can I talk to... • Talk to... • Ask... • At... • Tell... • Tell me...
Deep links “Talk to service about topic”
Not deep links “Talk to service.”
Conversations You need to end them
API.AI Actions SDK
API.AI Actions SDK GUI
API.AI Actions SDK GUI Natural Language Understanding
API.AI Actions SDK GUI Natural Language Understanding Integrations
API.AI Actions SDK It’s all on you GUI Natural Language
Understanding Integrations
None
$ API.AI Agent Entities Actions
Intent examples • Set an alarm for 6:00am • Wake
me up at 6:00am • I need an alarm at 6:00am • Set an alarm for March 1st at 6:00pm
Intent examples • Set an alarm for 6:00am • Wake
me up at 6:00am • I need an alarm at 6:00am • Set an alarm for March 1st at 6:00pm
Intent examples • Set an alarm for 6:00am • Wake
me up at 6:00am • I need an alarm at 6:00am • Set an alarm for March 1st at 6:00pm
Intent examples • Set an alarm for 6:00am • Wake
me up at 6:00am • I need an alarm at 6:00am • Set an alarm for March 1st at 6:00pm
Welcome Intent Talk to Clocky
Fallback Intent Tell Clocky to order me a pizza
Fallback Intent Tell Clocky to order me a pizza I
don’t know how to help with that.
Action Capture parameters
Action Capture parameters | Prompt for parameters
System Entitities • Number • Name • Location • Date/time
• Airport • And so much more!
Custom Entities • Food menu • Team names • Fantasy
names
Response These are logic-less
Demo time
Advanced Fulfillment For everything else, there’s webhooks
$ (Your action) Input Action Response JSON Request JSON Response
$ Your server
$ Your server API.AI Webhook
$ Your server API.AI Webhook (also JSON)
Hosting Host anywhere
Hosting Host anywhere | Node.js recommended
Simple webhook server let Assistant = require('actions-on-google').ApiAiAssistant; let app =
express(); app.post('/', function (req, res) { const assistant = new Assistant({request: req, response: res}); } }
Basic webhook let actionMap = new Map(); actionMap.set(”set_alarm", setAlarm); assistant.handleRequest(actionMap);
function setAlarm() { assistant.tell("Hello from the webhook!"); }
Basic webhook let actionMap = new Map(); actionMap.set(”set_alarm", setAlarm); assistant.handleRequest(actionMap);
function setAlarm() { assistant.tell("Hello from the webhook!"); }
Webhook with arguments function withArguments() { let time= assistant.getArgument(”time"); assistant.tell(”I
will set an alarm for" + time); }
SSML- Speech Synthesis Markup Language function withSsml() { assistant.tell("This response
uses” + "<say-as interpret-as='characters'>SSML</say-as>.” + "Isn't it <break> awesome?"); }
User Identity • ID (resettable!) • Display name • Given
name • Family name • Coarse device location (zip code and city) • Precise device location (lat/lng, address)
User Identity • ID (resettable!) • Display name • Given
name • Family name • Coarse device location (zip code and city) • Precise device location (lat/lng, address) Requires permission
Context Context allows for natural conversations
What’s the weather like in Minneapolis? It is currently 10°
in Minneapolis. !
What’s the weather like in Minneapolis? It is currently 10°
in Minneapolis. ! Context Location: Minneapolis Temperature: 10°
What’s the weather like in Minneapolis? It is currently 10°
in Minneapolis. ! Context Location: Minneapolis Temperature: 10°
Interaction 1 Interaction 2 Context Output Input
Output Context Contains all parameters from the action
Input Context Intents only trigger if context exists
Input Context Intents only trigger if context exists Allows for
branching flows
Context - Lifetime Duration (in requests)
Context - Lifetime Duration (in requests) | Ends with conversation
Demo time
Publishing
Actions Directory
Create a Google Cloud Project Register the Action Deploy via
API.AI Wait for review
Create a Google Cloud Project Register the Action Deploy via
API.AI Wait for review
Create a Google Cloud Project Register the Action Deploy via
API.AI Wait for review
Create a Google Cloud Project Register the Action Deploy via
API.AI Wait for review