🧝🏼 🥑 🤒 How we developed a visual designer for training digital agents 👩🏻‍✈️ ⚒️ 🎄

Hello again, Habr! In the last article, we talked about how we create a voice agent (we prefer this term, not a "voice robot") on only 500 records. Briefly recalling the content, we have learned how to successfully train agents on a very small base of calls.

To make it clear why this is - our company provides contact center services based on artificial intelligence. Well, to build the logic of the dialogue, we use a visual designer, which increases the efficiency of this work. In the new article, we are just talking about what kind of system is and how it works. With screenshots and cases - whatever you like.

How it all began

Like any other company, Neuro.net and our technologies are developing gradually. Our digital agents have been great for a few years now. But their training was very difficult.

In the previous version of the platform, we set the dialogue script using a standard set of actions, the order of which was determined by the rows in the table. The dialog tree was built using different conditions and a goto action. We have grouped actions into logical units.