“So many of our dreams at first seem impossible, then they seem improbable, and then… they soon become inevitable.”

- Christopher Reev

Dream: A Powerful Word

Imagine AI assistants being capable of understanding us, and talking to us. Imagine them learning from us and teaching us. Imagine them being our trusted assistants. Imagine them doing everything we want. Imagine them empowering us to develop personally.

Today, we make a very early version of our DREAM accessible to everyone. DeepPavlov DREAM is an AI assistant based on the socialbot built by our team for the Alexa Prize 2019 competition. Currently, it is available on our demo website and in Telegram messenger. DREAM blends together almost 40 different chit-chat and task-oriented skills to engage in open domain conversations. It relies on a selection of modern NLP models and components including 14 annotators, 4 post-annotators, and knowledge graph integration. Please, chat with DREAM and give us feedback on how to improve it! Don’t forget that it is still in the early stage and might be confused easily.

The DREAM is created on top of the DeepPavlov Agent, an open source multiskill orchestration framework. In an accompanying blog post we will guide you through the very simple DP Agent configuration with only one custom skill and built-in skill and response selectors. Read it to learn how to build AI assistants using our technology.

If you want to dive into details check DREAM socialbot Technical Report for Alexa Prize 2019 competition.


Does it mean that the dream of smarter and smarter assistants has to be postponed?

DeepPavlov addresses this challenge by introducing modular architecture for conversational agents. In the DP Agent framework, functionality for distinct tasks is packed in separate conversational skills with clear interfaces. This additional level of abstraction allows us to add new skills without interference with existing components of the system. Dialog as a whole is controlled at both individual skills and overall dialog levels. MVP for a new task can be tested in isolation before integration into an AI assistant.

Moreover, many skills such as chit-chat, alarm, calendar, etc. are common for the majority of use cases. So, why reinvent the wheel again and again? DP Agent makes it possible to create a conversational agent distro which includes a set of essential default skills. Default assistant provides basic functionality out of the box and can be extended by plugging domain specific skills.

Life-cycle of AI Assistant

DP Agent open architecture is especially powerful for building and maintaining complex conversational solutions. Integration of skills as micro services makes an assistant scalable. Development and support of skills can be effectively performed by a group of enthusiasts, distributed product teams, or subcontractors.

DREAM is a first multiskill distribution and an experimental AI assistant built with DeepPavlov conversational AI stack. In the coming months we will start to open source code for DREAM skills, services, and tools. We’ll also continue posting new blog posts to guide you through building more complex AI assistants using DREAM. Our roadmap includes publishing sample configs, tools for multiskill assistant design and dependency management. We plan to support fluid form-filling, multi-intent understanding, context tracking, learning from users, and many other critical scenarios in the future.

Announcing DeepPavlov Contributor Program

We already have some inspiring stories of contributors to our DP Library, and we welcome you to learn more about the program here.

Announcing Community Calls and State of the Union Events

Starting with this month, we will hold regular Community Calls, and every half of the year we will hold State of the Union events, to keep you informed of our developments, as well as to bring your feedback to us as soon as possible.

We will also keep posting updates in our DP Blog, and send bi-weekly newsletters to our subscribers. You can subscribe to our news here.

Dream Now!

Authors: Mikhail Burtsev, Daniel Kornev