Rasa provides an open-source conversational AI platform that is unique in its ability to use machine learning to learn dialog patterns from example dialogs. Key Rasa characteristics include:
- Open platform with vibrant and rapidly growing developer community (over 500K downloads)
- On-premise or cloud deployments possible.
- One of the only conversational platforms that has the ability to use machine learning to train the dialog
- Fully open and customizable NLU pipeline (versus the black box NLU engines of SaaS-type conversational platforms)
Conversational IVR with Rasa and VoiceXML
Thanks to our VoiceXML custom connector for Rasa, we can deliver Rasa-based conversational IVR solutions on any VoiceXML-compliant IVR platform. This includes all the leading IVR platforms on the market today, including Genesys, Cisco, Avaya, Aspect.
Why work with Nu Echo on your Rasa project
17 years of conversational expertise
Conversational user experience design is about how to ask questions to get predictable responses; how to handle user digressions, corrections, changes of mind; how to manage NLU confidence scores; how to recover from an ASR or NLU failure; how to lead the conversation to maximize success while giving users a sense of control. It is both art and science. But it’s certainly not the kind of expertise that can be improvised. It’s built on years of experience developing best practices and learning what works and what doesn’t.
We were doing conversational user experience design long before people even knew it existed. We have built our conversational practice over the past 17 years so that you don’t have to repeat the mistakes of others.
Proven engineering practices
No matter what development tools you use, an intelligent virtual agent is a complex application that needs to be developed with solid engineering practices that enable you to frequently release new versions that are always bug-free and perform flawlessly.
Our extensive Rasa tooling facilitates implementation of conversational AI solutions and ensures they have the expected behavior. This includes:
- Action unit test framework – A unit test framework for actions
- End to end test framework – Test a deployed application using scenarios to ensure all the solution parts are working properly together.
- Custom fallback action – While Rasa provides some fallback actions, we provide an implementation that tracks the number of errors and provides progressive error messages and transfer support.
- Generic help action – provides contextual informational messages.
- Low confidence confirmation policy – Allow the user to confirm or correct elements that the ASR recognized with insufficient confidence.
- Change policy – Sometime you just want to change your mind, this Rasa policy allow you to modify any information specified in a form.
VoiceXML IVR support
We have chosen Rasa as our conversational platform for VoiceXML IVR applications. To make this happen, we developed two key Rasa extensions:
A VoiceXML custom connector that makes the Rasa application compatible with all VoiceXML compliant IVR platforms.
A VoiceXML NLG server, that takes care of packaging all speech recognition grammars and audio prompts. This NLG server offers several important features, including:
- Support for multilingualism
- Structured message definition files (YAML)
- Audio file support and concatenation templates
- Sequential & random alternatives
- Resource integrity report
NLU optimization expertise: Because accuracy matters
No matter how good a conversational design is, if NLU accuracy is not there, the entire virtual agent project will end up in failure. For the past 17 years, we have been refining our NLU optimization methodologies and tools to ensure that our conversational solutions deliver the best possible accuracy, success rate, and user experience.
In fact, that’s what we are best known for. Companies across North America come to us simply to improve NLU and ASR accuracy of their current conversational applications.
Get in touch
The platforms and technologies you use make all the difference. We can help. Fill out the form below and we’ll get back to you soon.