janis - SDK For Human + AI Conversational Experiences

For Bots Built in Ruby

Bots enable businesses to respond to customers immediately but they often fail to understand user intent. According to Facebook, bots fail 70% of the time. janis helps solve this problem with a toolkit to easily keep humans in the loop when AI fails your customers. The solution includes an SDK for bot developers to connect their bots to Slack, and a Slack app to get alerts, then pause and take over a bot.

Solution

You can integrate janis in minutes and it begins working immediately, enabling you to deliver exceptional human + AI conversational experiences. This module has been tested with Messenger, Slack, Skype, and Microsoft Webchat. Please see our examples. It supports bot developers working in Node, Python and Ruby.

What you can do with janis:

You can view a full list of features at (https://www.janis.ai). It's core purpose can be explained with this single GIF

Takeover

What you need to get started:

Operational Dependencies:

You'll need an API key from janis and for each Chatbot a Bot Token. You can get both of those (free) when you add janis to Slack and through a conversation with janis.
If you're building a Messenger Chatbot, you'll need to setup a Facebook App, Facebook Page, get the Page Access Token from Facebook and link the Facebook App to the Facebook Page for janis to work.

Installation

$ gem install janis

Usage

Set your environmental variables for JANIS_API_KEY, JANIS_CLIENT_KEY, ACCESS_TOKEN.

$ export JANIS_API_KEY=xxxxxxxxxxxxxxxxxxxx
$ export JANIS_CLIENT_KEY=xxxxxxxxxxxxxxxxxxxx
$ export ACCESS_TOKEN=xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Add the janis class to your code and set the required parameter values.

require 'janis'

# janis Api Key
janis.apikey = ENV['JANIS_API_KEY']
# Unique janis Client Key for your bot
janis.clientkey = ENV['JANIS_CLIENT_KEY']
# possible values: "messenger" or "slack"
janis.platform = "messenger" 
# Page Access Token (only required for Messenger bots)
janis.token = ENV['ACCESS_TOKEN']

Incoming Message Schema:

Throughout this documentation, you will see references to incomingMessage. Depending on whether you have a Messenger or Slack bot, the schema will be different. The value of incomingMessage should be equal to the message you receive directly from either the Messenger webhook response, or from the Slack RTM event response.

# Example of a Slack Incoming Message
{
    "type": "message",
    "channel": "D024BE91L",
    "user": "U2147483697",
    "text": "Hello world",
    "ts": "1355517523.000005"
}

# Example of a Messenger Incoming Message
{
  "sender":{
    "id":"USER_ID"
  },
  "recipient":{
    "id":"PAGE_ID"
  },
  "timestamp":1458692752478,
  "message":{
    "mid":"mid.1457764197618:41d102a3e1ae206a38",
    "seq":73,
    "text":"hello, world!",
    "quick_reply": {
      "payload": "DEVELOPER_DEFINED_PAYLOAD"
    }
  }
}

Outgoing Message Schema:

Throughout this documentation, you will see references to outgoingMessage. Depending on whether you have a Messenger or Slack bot, the schema, as defined by each platform, will be different. Every time you track an outgoing message, the schema requirements match the respective platform.

# Example of Slack Outgoing Message
{
    "channel": "C024BE91L",
    "text": "Hello world"
}

# Exmaple of Messenger Outgoing Message
{
  "recipient":{
    "id":"USER_ID"
  },
  "message":{
    "text":"hello, world!"
  }
}

Tracking received messages:

When your bot receives an incoming message, you'll need to log the data with janis by calling to janis.hopIn. Note: janis can pause your bot so that it doesn't auto response while a human has taken over. The server response from your hopIn request will pass the paused state. Use that to stop your bot from responding to an incoming message. Here is an example:

hopInResponse = janis.hopIn(incomingMessage)
if hopInResponse['paused'] != true
# proceed to process incoming message
 ...

Tracking sent messages:

Each time your bot sends a message, make sure to log that with janis by calling to janis.hopOut. Here is an example of a function that we're calling sendIt that tracks an outgoing message and at the same time, has the bot say the message:

def sendIt(channel, text)
    # schema matches Messenger
    outgoingMessage = {recipient: {id: channel},message: {text: text}}
    janis.hopOut(outgoingMessage)
    client.say({'text': text, 'channel': channel})  # <= example of bot sending reply
    ...

Log Unknown Intents:

Find the spot in your code your bot processes incoming messages it does not understand. Within that block of code, call to janis.logUnkownIntent to capture these conversational ‘dead-ends’. Here's an example:

# let the user know that the bot does not understand
sendIt(recipient_id, 'Huh?')
# capture conversational dead-ends.
janis.logUnknownIntent(incomingMessage)

Dial 0 to Speak With a Live Human Being:

janis can trigger alerts to suggest when a human should take over for your Chatbot. To enable this, create an intent such as when a customer explicitly requests live assistance, and then include the following lines of code where your bot listens for this intent:

# match an intent to talk to a real human
if text == 'help'
    # let the user know that they are being routed to a human
    sendIt(recipient_id, 'Hang tight. Let me see what I can do.')
    # send a janis alert to your slack channel
    # that the user could use assistance
    janis.assistanceRequested(incomingMessage);

Human Take Over:

To enable the ability to have a human take over your bot, add the code below to subscribe to the 'chat response' event. Alternatively, if you'd prefer to use a webhook to receive the payload, please get in touch with us at [email protected] and we can enable that for you.

# Handle forwarding the messages sent by a human through your bot
janis.on :'chat response' do |data|
    text = data['text']
    channel = data['channel']
    client.say({'text': text, 'channel': channel})  # <= example of bot sending message
end

Go back to Slack and wait for alerts. That's it! Be sure to check out our examples.