What's new in January 2020

Tab completion, Autoflows, multi model serving and faster training

Tab completion

With the TAB key agents autocomplete their sentences, with Deepdesk suggestions. Write faster! 🦾

Smart compose

The suggestions appear as grey text in the input field. By hitting tab, the text is added to the message. Tab completions show either the common part of multiple suggestions, or the whole sentence if only one suggestion is available.

Future applications

This also paves the way for alternative methods to suggest parts of frequently used sentences. Either coming from the same recommendation engine, or from new sources like Natural Language Generating models.

Deepdesk - Tab completion.gif

Autoflows

With Autoflows, agents can launch a predefined script that will handle a specific (mini) conversation with a customer.

Automate repetitive conversations 🔁

It can be used for frequently occurring exchanges, like dealing with login issues. The script sends a number of questions to the customer. The customer can answer with text and or the quick reply buttons.

Hands-off 🙌

As long as the conversation follows the predicted flow, no action is required by the agent. If the flow is interrupted or aborted the agent can take over right away.

Autoflows eliminate manual repetitive work for the agent, and a quick, uninterrupted exchange with the customer. Bots done right.

Faster Training

Deepdesk training pipeline now runs on GPUs 🏎💨

Due to the growing amount of data and increasing complexity of the machine learning algorithms, training* times were sometimes taking up to 60 hours.
This was starting to limit further optimisation and hampering research and development experiments on larger data sets.

8 x faster

The model training pipeline was rebuilt to support GPUs for faster training. The new clustering and training pipeline, running on Tensorflow, is now able to run within hours. 

Deepdesk - GPU training.png
Multi model Deepdesk Dimelo.png

Multi model serving


The Deepdesk integration in Dimelo is now able to serve multiple models. The model served is determined by the categorisation of the conversation, like chat vs messaging or multiple brands within the same domain.

More?

Read more in our November 2019 post 🤓

Header image by Jude Beck on Unsplash