This capsule is the the companion code to the Bixby Space Resorts Sample Capsule guide. Space Resorts is a fun capsule that allows you to book a space vacation! You can find a space resort, book a reservation, look up a reservation, and change or cancel a reservation. This advanced capsule demonstrates development and design of searches, transactions and UI.
You can see all trained utterances and plans by entering this query in the
training tab search bar: goal:SpaceResort#all -has:continue. Examples:
-
Find space resorts
-
Show me space hotels with crater canyoneering
-
Search for hotels around The Red Planet
-
Look for space hotels with quantum bungee jumping around Saturn
We train these to have the goal SpaceResort#all and annotate any resort names,
planets and search criteria as Values. Here SpaceResort#all is a property
projection to the all property of the SpaceResort. This property is a
boolean of type ViewAll, which is always true and is used as a proxy to
signify that we want to display the full space resort, instead of focusing on a
single property like gravity or planet (see property projections below). A
match-pattern ties the views and dialogs for this property projection to
describe the whole hotel.
match {
ViewAll(all) {
from-property {
SpaceResort (result)
}
}
}
Why treat this as a property projection instead of setting the goal to the
SpaceResort Structure?
When the user wants to book a space resort and there are many possible
candidates in context, the user picks a single one using a selection prompt, in
order to continue with the booking (see SpaceResort Selection Prompt selection
below for full details). The context for that prompt is SpaceResort, and
selection prompt training must always have the same goal as its context.
Accordingly, selection prompt training for SpaceResort needs to use
SpaceResort as its goal.
The selection prompt training also uses a special flagged signal to route the
plan through the SelectResort action that filters the hotels currently in
context based on the newly provided inputs. It is crucial to add the
SelectResort flagged signal to the selection prompt training annotations to
achieve this behavior. However, we do not want to add the SelectResort
flagged signal to the "find" queries because these should issue a new search via
the FindSpaceResorts action instead of filtering existing results via the
SelectResort action. Since the selection prompt requires different annotations
patterns compared to the "find" queries, and annotation patterns must be
consistent for the same goal, this means that they must use a different goal.
This example demonstrates two points. First, the selection prompt
training must use SpaceResort as a goal. Second, that the "find" queries
must use a different goal than the selection prompts. By putting these together,
we deduce that the "find" queries cannot use SpaceResort as a goal. Therefore,
we use a distinct goal (SpaceResort#all) for the "find" queries in order to
provide consistent annotation patterns per goal.
Why don't we set the goal to the FindSpaceResorts action?
Another alternative would be to set the goal to the specific FindSpaceResorts
action, making it very clear how to fulfill the request. This approach would
simplify the match patterns for views, so we would not need to use the
from-property key. However, we use the property projection approach
so that our final resting context for "find" queries is the same as for property
projection queries. This means that we can pivot between all these states
seamlessly, or launch the "book" flow from any of these. Example conversation:
-
Find space resorts
-
The second one
-
What planet is it on?
-
What's the gravity there?
-
Book it
You can see all trained utterances and plans by entering this query in the
training tab search bar: goal:SpaceResort#all has:continue. Examples:
-
On Jupiter
-
With low gravity
-
Only the ones that are kid-friendly
These are continuations of the outer "find" queries that allow users to refine
their space resorts search by providing additional inputs. Since the goal for
outer "find" queries is SpaceResort#all, we annotate both the goal and the
"Continuation of" to also be SpaceResort#all. Any resort names, planets and
search criteria are annotated as Values. This reissues a search with the new
inputs being added to those already in context.
You can see all trained utterances and plans by entering this query in the
training tab search bar: goal:SpaceResort#* -goal:SpaceResort#all -has:continue. For example:
-
What's the gravity at The Mercurial?
-
Where is Io-Tel?
Here, the user is asking to know about a specific property of a space resort,
such as the gravity or the planet. We train the goal to be that property
projection (ex: SpaceResort#gravity), and we annotate any resort names,
planets and search criteria as Values. We also add a special flagged signal
route to ProjectResort. This is in case there were multiple space resorts
that matched the search inputs. Then the ProjectResort action will ask the
user to select a single space resort before providing the answer.
You can see all trained utterances and plans by entering this query in the
training tab search bar: goal:SpaceResort#* -goal:SpaceResort#all has:continue. For example:
-
What's the gravity there?
-
What planet is it on?
We train these just like the outer property projections, with the addition of a
"Continuation of" SpaceResort. This allows pivoting between inner/outer "find"
queries and inner/outer property projections.
You can see all trained utterances and plans by entering this query in the
training tab search bar: goal:MakeReservation -has:continue. For example:
- Make a reservation for a space resort on Mars the third weekend in December for 2 astronauts
We train these to the goal MakeReservation, which is the Action to finalize the
transaction. We also add two flagged signal routes: CreateItem and
CreateOrder. We annotate as Values any present inputs for either "find" or
"book", such as resort name, planet, search criteria, number of astronauts, etc.
This creates a plan to first find a space resort that matches the search inputs,
then prepare an Order and pass it to the MakeReservation action, which will
present the user with a Confirmation screen to review and agree to the
reservation.
You can see all trained utterances and plans by entering this query in the
training tab search bar: goal:MakeReservation continuation:SpaceResort. For
example:
-
Make reservation
-
Book a pod for 2 astronauts
We train these just like the outer "book" queries, with the addition of a
"Continuation of" SpaceResort. This is to cover cases where the users are
already browsing space resorts and want to initiate a booking for one of the
results in context.
You can see all trained utterances and plans by entering this query in the
training tab search bar: goal:MakeReservation continuation:MakeReservation. For example:
-
Pick a different habitat pod
-
Change that to 2 astronauts
-
Select a different date
We train these as "Continuation of" MakeReservation for cases where the
users are at the Confirmation screen to review their order and decide that they
want to make some changes. The goal remains MakeReservation, and any Values
are annotated as such (ex: number of astronauts, pod name). This time, the
flagged signal route is to ChangeOrder. This is to re-route the request to
update the Order with the newly provided information. For generic requests that
do not contain a new input Value (ex: Change the number of astronauts), we
add an extra flagged signal route to the action for that request (ex:
GetNumberOfAstronauts).
You can see all trained utterances and plans by entering this query in the
training tab search bar: prompt:Confirmation. For example:
-
Yes
-
Do it
When the user is done reviewing their Order at the Confirmation Prompt, they can
use these utterances to move forward and proceed with the reservation. This is
an "At prompt for" Confirmation with goal Confirmation. The Confirmation
itself is annotated with a boolean Value "true" or "false".
You can see all trained utterances and plans by entering this query in the
training tab search bar: prompt:SpaceResort. For example:
-
The one with a refueling station
-
The Mercurial
-
The one on Venus
The booking flow only allows a single SpaceResort at a time, so when there are
multiple candidates, the user will be presented with a SpaceResort Selection
Prompt. For prompt training, the goal must always match the prompt context, so
we train these as "At prompt for" SpaceResort with goal SpaceResort. There
are many ways the user can answer, so we annotate any provided Value (space
resort name, planet, search criteria) and add a special flagged signal route to
the SelectResort action. This action will take the hotels currently in
context and attempt to filter them based on the newly provided inputs. For
example:
-
User: Book a hotel near Jupiter
-
Bixby: Here are some hotels around Jupiter. Which one would you like?
-
User: The one with a refueling station.
-
Bixby: There are 3 hotels around Jupiter with a refueling station. Which one? (Where the 3 options are a subset of the previous options, not a new search)
-
User: The Ganymede Moon Motel
-
Bixby: (Proceeds with the booking flow)
You can see all trained utterances and plans by entering this query in the
training tab search bar: has:prompt -goal:SpaceResort -goal:Confirmation. For
example:
-
3 astronauts
-
The Honey Moon Suite
-
Next weekend
For prompt training, the goal must always match the prompt context, so we train
these as "At prompt for" <Concept> and goal <Concept>. Then we annotate the
Value in the utterance for that <Concept>. For example, "At prompt for"
HabitatPod has goal HabitatPod and the PodName is annotated as a value.
- Bixby Developer Center - Everything you need to get started with Bixby Development!
- Quick Start Guide - Build your first capsule
- Design Guides - Best practices for designing your capsules
- Developer Guides - Guides that take you from design and modeling all the way through deployment of your capsules
- Bixby Developers YouTube Channel - Tutorial videos, Presentations, Capsule Demos and more
- Bixby Developers Chat - Voice, Conversational AI and Bixby discussions
- @BixbyDevelopers - Twitter
- Have a feature request? Please suggest it in our Support Community to help us prioritize.
- Have a technical question? Ask on Stack Overflow with tag “bixby”