Chatbot | Web Widget | Mobile App

• Upgraded for Rasa 2.6.0.

• Upgraded for Rasa 2.5.0:
  • Problems:
    • Rasa 2.4.0: ‘charmap’ codec can’t decode bytes in Arabic and Armenian. Check rasa topic #41569, rasa issue #8215, rasa pull #8286.
    • Rasa 2.4.2 fixed above problem but training with checkpoint_model: True in the pipeline fails. Check rasa topic #41706 and rasa issue #8296.
  • Supports SpaCy 3.0.
• Pipeline & Policies:
  • Now uses LanguageModelFeaturizer in pipeline.
• Actions:
  • Refactored and added helper functions:
    • Added SlackApp class to send messages to Slack from custom actions.
    • DatabaseConnection constructor can now take optional parameters (either hostname, database, username, password, or a list of them).
    • get_text_from_lang can now take a list of lists instead of list of strings to choose a random answer.
  • ValidateFormTroubleshootInternet now connects to a database to get the required slots.
  • ActionRequestHuman sends message to Slack with username, phone number, Rasa ID, and slot values.
• Intents & training data:
  • Added request_human intent which starts action_request_human.

• Upgraded for Rasa 2.3.4:
  • DIET, ResponseSelector, and TED's model_confidence went back from cosine to linear_norm (rasa issue #8014).
  • Studying Rasa 2.4.0.
• Pipeline & Policies:
  • See modifications mentioned above.
  • Tensorboard integration.
  • Other minor changes.
• Actions:
  • Refactored and added helper functions:
    • Language helper functions now take a list as argument instead of one parameter per language. This helps easily scale for any number of languages.
    • get_lang() and get_lang_index() are used in the get_*_from_lang() functions to avoid replication of code.
    • Added reset_slots() helper function.
    • Prettified output of announce().
  • Override ActionSessionStart default action. Now slots are forgotten after a session starts.
  • ActionAskSlotNames in FormValidationActions now have a "stop" (🚫) button to stop the form.
• Intents & training data:
  • Added internet_problem intent which starts form_troubleshoot_internet form.
  • Removed Pokémon features.
• Forms:
  • Added form_troubleshoot_internet with its FormValidationAction and ActionAskSlotNames.
• Multilanguage support:
  • Chatbot now also understands Lebanese texting.

• Actions:
  • Added functions (Modified Database class & language-dependent template).
  • Converting utterances into actions that utter the message corresponding to the selected language.
  • Minor code changes.
• Intents & training data:
  • Working on basic customer support.
• Multilanguage support:
  • Converting utterances into actions that utter the message corresponding to the selected language.

• Actions:
  • Added helper classes and functions (Database query & language-dependent utterance)
  • Minor code changes.
• Multilanguage support:
  • Chatbot understands 4 languages: English, French, Arabic, Armenian.
  • Instead of utterances we should use actions that utter the message corresponding to the selected language.
  • The chatbot will reply in a select language independently of the language of the input. The user can change this language at any time.

• Actions & training data:
  • Queries database when asked about quota (using form).
  • Added how_are_you intent.
• Other:
  • Added lookup table for person names.

• Upgraded for Rasa 2.3.0:
  • DIET and TED confidence use now cosine distance.
  • DIET loss type is now cross_entropy (softmax is deprecated) and constrain_similarities is set to True (recommended when using cross_entropy).
• Actions & training data:
  • Weather action now supports inputs without a city name and will use last city name in that case.
  • A new out_of_scope intent was added with a corresponding custom action that does a Google search.
• Pipeline:
  • See Rasa 2.3.0 modifications mentioned above.
  • The pipeline was changed to use SpaCy. The chatbot does a way better job at detecting city names.
  • Number of epochs was overkill. It went from 16k to 128 epochs for DIET and TED.
• Other:
  • Added tensorboard support.
  • lookup entry in nlu.yml is removed, added data/lookups/pokemon_name.yml instead to let Rasa search automatically.

• This chatbot can recognize if a given Pokemon exists and gives you the weather in a given city or country.
• Continuous Integration support, Non-Countinuous Deployment via docker push/pull.