Want to Improve Your AI Assistant? Then You Need Rasa X

Update: A lot of things have changed since this page was written. Rasa X, the freemium companion tool to Rasa Open Source, is no longer supported or maintained, and we are currently focused on the development of the Rasa Enterprise platform. To learn more about this, you can check out this blog post.

Want to Improve Your AI Assistant?
Then You Need Rasa X

Developing an AI assistant involves many steps: finding or creating training data, sorting through many conversations, keeping track of changes to your models, to name a few. And after doing all that work, improving your assistant is an ongoing process. Maybe your assistant is sometimes thrown by unexpected input or has trouble handling certain conversation patterns. Maybe you want to identify which new intents you should support.

You need input from the real world before you can improve your assistant successfully.

And that is why we built Rasa X.

What is Rasa X?

Rasa X is a browser-based tool for developers that layers on top of Rasa Open Source. Our UI tool makes it easier to improve the quality of Rasa-powered assistants because it allows you to:

View and filter conversations between your assistant and users.

Convert conversations into training data.

Easily give test users access to your assistant.

Manage and version assistant models and training data.

Rasa X is not a hosted service; it can be deployed anywhere.

Your training data and conversations are never passed to Rasa, so they are kept private and secure. We offer two plans for Rasa X: free and enterprise (paid).

Why Do You Need Rasa X?

Because you need good training data

Quality training data is hard to come by and even harder to create yourself. You need training datasets for your specific domain and use case. But most of the large, publicly available conversational datasets available today are general purpose. Some developers spend weeks or months creating training data themselves from scratch. But it’s impossible to anticipate all the ways users communicate with an assistant.

You need datasets that reflect the vocabulary and dialogue specific to the domain and tasks of the assistant. For example, a health insurance assistant needs to know how to answer questions regarding the status of claims and insurance payments. A travel assistant needs to understand queries about hotels, flights, and restaurants. Want your assistant to tell a user their bank account balance? It needs to understand language specific to banking.

The best training data comes from real conversations from real users. Training data from generic datasets or hypothetical conversations leads to poor user experiences.

Recent work by Facebook AI Research found that:

"Dialogue research today functions almost entirely based on extensive supervised learning from humans talking to one another — usually crowdsourced or publicly available on the internet. This data can differ significantly in distribution from the environment in which a chatbot might be deployed. To help researchers further explore and push dialogue research forward, it’s important to have agents out in the real world actually conversing with humans."

To get good training data, you need Rasa X

With Rasa X, you can get good training data for your assistant quickly and easily. Rasa X converts conversations—collected from test conversations and real user interactions—into quality training data. The training data can be used to re-train and improve your assistant.

You can use Rasa X to add example utterances to training data and correct intents that were misclassified. And while you could annotate training data using the command line, Rasa X allows you to annotate large amounts of training data far more efficiently.

Because you need to sort through datasets quickly

Improving your assistant requires sorting through datasets to identify the conversations that matter. The data collected from interactions between the assistant and users is a goldmine of insights that you can use to:

Identify which tasks the assistant handles successfully or unsuccessfully.

Figure out which new skills or features to prioritize based on usage.

Improve your assistant based on user interactions and behavior.

Many developers export conversations to a spreadsheet, which adds another step to their workflows. Spreadsheets make it difficult to analyze conversation data meaningfully because the files can be enormous, and extracting insights is a manual task. Manually collecting, sorting, and tagging conversations is a lot of work and takes time. And once your assistant starts having hundreds, or possibly thousands, of conversations, it will be difficult to process every conversation manually.

Rasa X makes it easier to gain insights from conversational datasets

Rasa X is a tool purpose-built for leveraging conversational data sets. With Rasa X, you can filter conversations based on criteria such as:

Length of the conversation.

Specific intents, e.g. out of scope.

Specific actions, e.g. fallback or human handoff.

You can tag conversations based on a variety of things such as:

Negative or positive user sentiment.

Whether the conversation has been reviewed or needs to be reviewed.

If the user got stuck while talking to your assistant.

If the user achieved their goal.

Because you need to follow DevOps best practices

Most development teams have embraced DevOps and follow DevOps best practices, which include using version control, continuous integration and continuous delivery (CI/CD), and automated deployment and testing. The machine learning world is now embracing the same set of practices for conversational AI applications. To be practical for product teams, tooling and workflows for building assistants must work with DevOps best practices.

Rasa X helps you follow DevOps best practices

Rasa X features integrated version control, which lets you connect your Rasa X instance to a remote git repository, like GitHub, Bitbucket, and GitLab. You can review and test all changes before they go to production. And you can keep track of all changes to your training data and models.

With Rasa X, you can enact a strategy of continuous improvement. You can feed your model training data based on the real conversations of users and automate quality assurance (QA) to improve your AI assistant continuously.

When Should You Use Rasa X?

You’ll get the most out of Rasa by using Rasa Open Source and Rasa X together.

Build your assistant with Rasa Open Source first. When your assistant is at the point where it can have basic conversations that achieve its main purpose, you should start using Rasa X.

Take your assistant to the next level with Rasa Open Source and Rasa X.