It's only been 9 months since we launched our first open source product, Rasa NLU, and we're super stoked that our awesome community has hit some important milestones:
30,000+ downloads and thousands of active developers
We knew there was demand for an open source NLU library from talking to developers at our #BotsBerlin meetup: most people we spoke to were uncomfortable relying on a third party SaaS solution for such a core piece of technology, but didn't have the time or resources to build their own. But we were still completely blown away by the reaction to Rasa NLU - now with over 30,000 downloads and thousands of active developers. We're seeing more and more developers moving away from tools like API.AI to Rasa NLU for full control over their data sets, lower latency, and higher customization in many different segments:
From the obvious use cases in customer service to others like sales and internal processes
300+ members on Gitter
Our community would not have been able to grow so fast without all the helping hands on our Gitter community channel. This is the place where newbies meet experienced Rasa devs and help each other. It's also the place where we collect feedback from the community and understand how developers use the product.
An open source project would not be anything without contributors. Our ambition is to build the best toolset for developers to enable truly sophisticated conversations. We've seen a lot of excellent contributions that make it easier to run Rasa NLU in a variety of environments, improving our docker builds, organising models into projects, and adding CORS support. That's why we're super excited that community members like @wrathagom, @PHLF, @vinvinod, and @paschmann actively support and contribute to Rasa NLU. @wrathagom also has an excellent series of blog posts on building a bot with Rasa NLU.
Rasa NLU started as a wrapper around open source NLP backends such as spaCy or MITIE to make it much simpler for developers to build their own NLU. Now it's evolved into a robust, scalable, thoroughly tested code base that's run in production by all sorts of companies. Through all the feedback from our community, we'll continue to invest into research and development to push the boundaries of conversational software. Particularly, we're focusing on two major topics:
(1) Dialogue Management: More and more community members start to realise the limitations of managing dialogue with a state machine and rule-based decision tree. We've been working on a new, machine learning based approach to dialogue management for over a year and are super excited to open source Rasa Core today.
(2) Next level NLU: NLU is by no means a solved problem, we love this paper for highlighting just how fragile many systems are. That's why we're excited that universities like TU Munich, TU Delft and the University of Edinburgh are using Rasa and pushing the boundaries of NLU. For example, we just finished a research project with character level models to make NLU less sensitive to typos.
We genuinely want to thank you all for your amazing support and commitment to Rasa. The last 9 months have been a great start but there is still a lot of stuff to be done to build truly conversational software - so let's get back to work!