RASA

"Open source machine learning tools for developers to build, improve, and deploy text-and voice-based chatbots and assistants". (cit. RASA github home page).

RASA is probably the most important open-source tool to develop "task-oriented" conversational applications. Despite the RASA official statement, the original project has not conceived to manage voice interactions, but just chatbots with some support to GUI/buttons.

RASA architecture consist in two main components:

• RASA NLU is based upon DIET algorithm, a a refined state of the art intent/entities "classifier
• RASA Core (now called RASA Dialog Manager), based on TED policy, a machine learning algorithm, to manage multi-turn dialogs, escaping the traditional state-machine based way, but instead allowing conversation developers to insert "stories", set of of intents-actions sequences (conversation examples). With end-to-end training, developers program the conversational agent dialog manager giving end-to-end turn-taking examples (the stories).

RASA owned in few years now, a huge open community of developers and researchers. It's probably the biggest open source project to develop on-premise "production-ready" complex dialog systems. All the development ecosystem is around the Python programming language.

References

• Home page: https://rasa.com/

• Github: https://github.com/RasaHQ/

• Community forum: https://forum.rasa.com/