• Sentences for defining intent should only contain the important core message. It is good to remove all unnecessary content and baggage (such as greetings, addresses, etc.) that repeats across different intents.

• Responses should be rather short (up to 20 words), if they are longer, they can be divided into separate sentences. This will result in a faster response (converting a long text to speech takes longer) and keep the communicating user's attention better.

• If there is another action included in the response, the user should be verbally informed, e.g. "Hold on, I'm calling now."

• In the case of follow-up scenarios, it is important to remember that the scenario mechanism is not contextual. Therefore, additional questions should always be formulated as open-ended to deduce the appropriate intent from the possible answer. For example, when introducing a company, I will not ask "Do you want to know more?" because a possible answer yes/no requires context, but I would rather ask: "Can I tell you more about our services or projects, ask what interests you?"

• We recommend grouping scenarios into categories according to their general areas of interest (Company, Navigation, ...) to help with content maintenance.

• Try to write responses consistently to correspond to the desired personality of the receptionist and maintain their integrity. Typically, it is not appropriate to switch between formal and informal language or alternate between decent expressions and the latest slang of adolescents.

• <break strength="weak"/> - a pause or a decrease in the voice with the indication of intensity - possible intensity values range from lowest to highest: "x-weak", "weak", "medium", "strong", "x-strong".

• <break time="200ms"/> - a pause for a specified time in milliseconds - suitable for longer pauses and dramatic pauses.

• <say-as interpret-as="ordinal">1</say-as> - says a number as an ordinal number: "first".

• <say-as interpret-as="characters">vut</say-as> - spells each letter separately. The same effect can be achieved without the tag by writing in capital letters: VUT or V.U.T.

• <say-as interpret-as="date" format="dmy">10-9-2021</say-as> - pronounces as the date "10th September 2021".

• <say-as interpret-as="time" format="hms">14:30:25</say-as> - pronounces as a time, i.e. "14 hours 30 minutes 25 seconds".

html
https://cloud.google.com/text-to-speech

javascript
[displayed text|pronunciation], for example, [ARTIN|artynn].

javascript
[this text will only be displayed, but nothing will be pronounced|].

javascript
[|this text will only be pronounced, but nothing will be displayed].

• $sys_date_time = date and time in the following format: Feb 2, 2021 14:42:22

• $sys_time = only time in the following format: 14:42:22

• $sys_month = month number in the following format: 2

• $sys_day_of_month = day number in the following format: 2

• $sys_day_of_week = day of the week number in the following format: 2

• $sys_12_hour_period = AM/PM flag in the following format: PM

• $sys_hour = current hour in the following format: 14

• $sys_minute = current minute in the following format: 42

• $sys_last_scenario = name of the last scenario in the format: Introduction scenario

• $terminal_sn = serial number of the terminal on which the current scenario is running (for differentiation if multiple terminals are using the same chatbot)

• {decline()} – example: decline($name, 4) – name declension

• {secondsfrom(timestamp)} – example: secondsfrom(1555683122.0) – number of seconds from the specified moment

• {date(locale, format)} – example: date(cs_CZ.utf8, %A %-d. %B) – date in the specified format

• {time} – current time in the format %H:%M

• {command=<USER_REQUEST>} – execution of a user request (see below) directly from the presented text, e.g. {command=EMAIL delivery Pavel Novak}

html
https://emojipedia.org/

Note: Dialogues are contextual "scenarios" that the user does not normally enter and are prepared for basic situations already in development.

• Receptionist's face

• Telephone services

• Video terminal

• Email interface

• Other services and devices connected within the integration of aReception for the customer

• VIDEOCONFERENCE - starts a video conference via VIDEOTEL

• CALL gradually - calls people from a pre-set list gradually (can be used as a backup option in case the user or back-office does not answer the phone)

• CALL <general | meeting | delivery | interview> <full name> - calls <full name> for a meeting | delivery | interview, for example: "CALL meeting Tomáš Ondráček"
• CALL {"type": "gradually | general | meeting | delivery | interview", "name": "value", "people2call": ["value1","value2"], "location": "value"} - JSON notation for the CALL command, allowing for an extended set of parameters (such as location or a list of contacts for gradual call).

• EMAIL delivery <full name> - sends an email about mail in the mailbox to reception for <full name>, for example: "EMAIL delivery Tomáš Ondráček" - email addresses must be available in the contacts list

• GET-CONTACT-INFO { "response-mapping": { "variable_name_without_dollar_prefix_1": "json_path_to_field_in_anwer_1", "variable_name_without_dollar_prefix_2": "json_path_to_field_in_anwer_2"}, “filter”:{"field_name": "filter_value_case_insensitive"}} - returns appropriate contact(s) from contact database from admin interface Return codes: "--GET-CONTACT-INFO-OK--” "--GET-CONTACT-INFO-ERROR--” Example: GET-CONTACT-INFO { "response-mapping": { "first_contact": "$.result[0].email", "all_contacts":"$.result"}, “filter”:{"phoneNumber": "778031466"}} Example of answer in JSON format with all field names: [{"id": 303, "note": "poznamka", "type": "zamestnanec", "email": "xxx@xxx.com", "office": "K125", "picture": "", "building": "D517", "idNumber": "123-987-AAD", "isActive": true, "language": "", "lastName": "Pavel", "location": "Ostrava", "firstName": "Jaroslav", "employment": "ekonom", "externalId": null, "phoneNumber": "+420111111111", "organizationUnit": "backoffice", "secondPhoneNumber": "+420999999999"}]

• END-CONVERSATION - ends the conversation (a conversation cannot be started for 10s - the person must have time to leave the camera)

• SOS - the user receives help (the same as when the question mark button is pressed in the bottom left corner)

• KEYBOARD <keyboard_type> - displays a keyboard for user touch input.
• KEYBOARD - displays a touch keyboard for selecting a name from a list of current contacts (the selected name is sent as a message, simulating keyboard input / representing speech).
• KEYBOARD contacts <text> - presets the <text> to the keyboard for name selection
• KEYBOARD general - displays a general keyboard for entering any text that is further processed as if the user had spoken it.

• SHOW-URL <url> - displays a window with a web page, the user can either close it themselves or it will automatically close when the user ends the conversation by leaving. The URL-CLOSED system intent is triggered when it is closed. The web page must be able to displayed in an iframe.
• SHOW-URL {"url": "http://example.org", "sandbox": "..."}
• the sendbox parameter is optional, if not specified or empty it is the most restrictive
• the allow-all value removes all restrictions
• individual restrictions can be overridden by the values described at https://www.w3schools.com/tags/att_iframe_sandbox.asp
• beware of allowing a link to appear in a new window that would overlay the avatar window
• TIP: most pages work well with the "sandbox": "allow-scripts allow-same-origin" setting.

• SHOW-IMAGE <url> - displays a window with an image loaded from the specified internet address, the user can either close it themselves or it will automatically close when the user ends the conversation by leaving. The IMAGE-CLOSED system intent is triggered when it is closed.

• READ-QRCODE <variable_name_without_$_sign> It will display an image from the camera's perspective and the user can point the camera at a QR code to load it. After reading the QR code, it is saved to a variable, the name of which is passed as a parameter of the command (without the initial $ character). Return values (to the chatbot): --QR-CODE-READ-- Successful QR code scanning --QR-CODE-CANCELLED-- User-initiated load interruption (cross in the top right) --QR-CODE-NOT-READ-- QR code reading timeout expired

• SWITCH-LANGUAGE <locale> - switches the language in which the receptionist communicates (this language must be in operation and therefore have the corresponding chatbot set up in that language - cs-CZ, en-US, etc.). Example: SWITCH-LANGUAGE en-US.

• WAIT <nnn> - waits for nnn milliseconds (e.g. WAIT 2000 waits for 2 seconds).

• WAIT-FOR-SPEECH-END - waits until the receptionist has finished speaking, then returns --SPEECH-FINISHED--. For example, if you want to display a web page (SHOW-URL) only after they finish speaking, otherwise it will display while they are speaking.

• ASK-CENTRAL-CHATBOT - calls the central conversational chatbot and asks it the same (last spoken) question or the input entered after this user request. The central chatbot then finds a suitable general answer instead of the specialized chatbot instance for a specific reception. The central conversational chatbot is continuously updated with new topics and conversation scenarios and can answer more general questions than the specialized chatbot instance for specific receptions. Typically, the central chatbot is called when the local chatbot does not know the answer to a question (i.e. in the system scenario "Unrecognized message"), but it can be used in other appropriate situations. This ensures that local, more specific answers are prioritized over general ones.

• ASK-CHATGPT <question> - submits a question to the artificial intelligence Chat GPT and displays its response. The parameter with the question can be obtained in the conversation by asking a question to the system entity @sys-input and saving it to a variable, which is then passed to the user request.

• INVOKE-CHATGPT-API <input> - calls the ChatGPT API and returns a response to the chatbot input. The response can then be retrieved in the @sys-input system entity. In case of an error, it will return --INVOKE-CHATGPT-API-ERROR--. An example follows:

json
INVOKE-CHATGPT-API {
"model": "gpt-4o",
"messages": [
{
"role": "system",
"content": [
{
"type": "text",
"text": "If the user input contains a destination, place or location, return that destination. If not, return NONE."
}
]
},
{
"role": "user",
"content": [
{
"type": "text",
"text": "$last_user_question"
}
]
}
],
"temperature": 0,
"max_tokens": 256,
"top_p": 1,
"frequency_penalty": 0,
"presence_penalty": 0
}

• GET-GPT-INTENT <text> - Can be used as an alternative to the NLP in the chatbot. Intents are defined in the administration application in the "GPT intents" section. The recognized intent is returned to the chatbot input. It can then be retrieved in the system entity @sys-input.

• LOG <text> - when the user request LOG is called, the terminal does not perform any specific action. The <text> value is visible in the chatbot's conversation history.

• RECOGNIZE-FACE - the user request performs face recognition of the person standing in front of the camera and returns a value:
• --RECOGNIZE-SUCCESS-- if the face was recognized.
• --RECOGNIZE-FAILURE-- if the face was not recognized.
The name of the recognized person, including the uncertainty value, is returned in the context variable $fr_user.

• SET-CHATBOT-INPUT <text> - sends the specified <text> to the chatbot input as if the user had spoken it into the terminal.

• MOOD <expression> <duration> <amplitude> - sets the facial expression (<expression>) for <duration> seconds, with an intensity of <amplitude>
• <expression> - ClosedSmile, OpenSmile, Sad, Angry, Fear, Disgust, Surprise, Thinking, Blush, LeftWink, RightWink, Blink, Scream
• <duration> - time, a whole number, in seconds
• <amplitude> - amplitude of the expression - typically 0-1, but some expressions can be exaggerated to 1.2 (it's a lot about trying it out because different facial expressions require different amplitudes)
• example: "MOOD ClosedSmile 3 1.1" - a closed-mouth smile for 3 seconds with an intensity of 1.1

json
HTTP-CALL <json-specification>

json
HTTP-CALL {   
  "method": <"POST"/“GET“/, …>   
  "url": <"service-url">,   
  "headers": {     <"header-key-1">: <"header-value-1">,     
                   <"header-key-2">: <"header-value-2">,     …   },   
  "payload": <“text-payload“>,   
  "request-timeout-seconds": <time-in-seconds>,   
  "return-chatbot-message": <message-to-send-to-chatbot-after-finishing-call>,   
  "response-mapping": {     
    <"context-var-for-resultCode">: "RESULT_CODE",     
    <"context-var-for-resultMessage">: "RESULT_MESSAGE",     
    <"context-var-1">: <"json-path-1">,     
    <"context-var-2">: <"json-path-2">,     …   
  } 
}

json
HTTP-CALL

{

"method": "POST",

"url": "https://8c5dae17-0b31-45f9-be49-c0263cd2b6f6.mock.pstmn.io/testpath?param1=val1&param2=val2",

"headers": {

"Content-Type": "application/json",

"X-auth": "kjnm45531"

},

"payload": {

"state": "idle",

"data": {

"welcome_message": "Dobrý den,",

"messages": [

"msg1",

"msg2",

"msg3"

]

}

},

"request-timeout-seconds": 10,

"return-chatbot-message": "—HTTP-CALL-FINISHED--",

"response-mapping": {

"resultCode": "**RESULT_CODE**",

"resultMessage": "**RESULT_MESSAGE**",

"contextParam1": "$.key",

"contextParam2": "$.license.test"

}

}