Mercury is the power tool for Conversational AI. The no-code platform requires no programming skills.
At Mercury.ai we believe that conversational AI is the key to user-centric services in our increasingly technological world. To make it generally accessible, we build our platform to overcome the dependence on hard-to-find AI developers and the technical insufficiency of drag and drop chatbot builders.
Mercury.ai gives business users a codeless interface to set up and maintain AI assistants and empowers developers to exploit the full potential of conversational AI through standard REST APIs. Because, for the user, AI should be exciting to see in action but easy to set up like any other business software. In that regard, it should not be different from a content management system, marketing platform, or CRM.
We are beyond excited that with today’s release, we are closing the loop for our fully managed, self-optimizing AI offering. After shipping lifecycle management for models and training data last year, then eliminating the need to manually tweak hyperparameters with the introduction of our Auto-ML NLU service a few months ago, today we introduce active learning capabilities.
From today on, all AI assistants on the Mercury.ai platform, in all pricing plans, will be able to learn from the user interaction intelligently and gather training data to extend their understanding and to improve their intent differentiation.
We are proud to present the fastest and most cost-efficient way for businesses to seize the full potential of conversational AI available today
Let’s quickly summarize what’s new and what you can use with your Mercury.ai account from today:
To improve its NLU a bot will actively gather new training data from user interactions. Whenever the bot has low confidence in the interpretation of a user message it will ask the user whether its interpretation is correct. If the user confirms the bot’s interpretation, the original user message will be used as new training data. While this continuously improves the NLU model this behavior also benefits the immediate user experience, since it provides a route for the user to achieve their original intent.
If user input can not clearly be resolved between to “rivaling” interpretations, the bot will now be able to actively disambiguate. Similarly to the active learning behavior, it will ask the user and then add the user message to the training data.
If at some point a conversation fails there is now the option to resort to a contextualized fallback message. You can now configure different fallback messages for different scenarios in the bot and thus send the right message to continue the conversation where it failed.
Get all intents that lead to conflicts in NLP and thus can negatively influence the speech understanding of the AI Assistant. On the project start page, you can see an evaluation for the live instance, in the training data view you can see it for each stage individually and correct the weak intents.
An NLU model is quite a dynamic beast with the constantly growing number of intents and training utterances. To understand how the quality develops over time and how new training data affects your model you can now see the history of your model and the most important quality metrics.
If you don't want to build your bots directly via the platform, but rather control everything via Code? No problem with our public API, which makes the functionality of the platform directly accessible via the REST API and webhooks.
New design, new minified mode, and, for all those enterprise legacy victims, it now supports that browser whose name we will not spell out.
With real-time information on the live bot. Find all suggestions for your training data and easy ways to resolve conflicts and the state of your project in general.
Our new and quite comprehensive documentation now provides in-depth info on the platform it’s concepts, features, and the API. Go check it out!