If you’re looking for a very accessible entry into the why’s and how’s of AI, conversational AI is a great starting point. In preparation for our upcoming Deep Learning Conference, I recently spoke with one of the conference presenters, Mady Mantha, Senior Technical Evangelist at Rasa.
Rasa builds the standard infrastructure layer for AI. Building contextual assistants and chatbots that really help customers is hard. Rasa provides infrastructure & tools necessary for high-performing, resilient, proprietary contextual assistants that work. With Rasa, all developers can create better text- and voice-based assistants. The company created Rasa Open Source, a free open-source machine learning framework to automate text- and voice-based assistants. It’s now used by millions of developers globally.
What’s particularly appealing is that Rasa is a tool that’s accessible ot everyone – even people without a degree in computer science or a deep understanding of AI. According to Mady:
“Developers that don’t know machine learning, that aren’t data scientists, product teams that may not be that technical can still take advantage of our products. Our mission is that AI shouldn’t be a black box. You can’t just package this up and say, well, we’re gonna hold all of this knowledge and no one else can know about it, and sit up in our ivory towers.”
Mady explained that Rasa’s users are mostly developers. “Right now there are front end developers, back end developers, DevOps engineers, just creating CI/CD pipelines and working on DevOps stuff. But it’s anyone who is somewhat familiar with Python, and with conversational AI, and they kind of want to build a chatbot, but they don’t know how and where to get started.”
About the Rasa community
“We like to nurture a community of makers. We have over eight million downloads, and three 3 million developers that love and use Rasa,” One of the great ways the community comes together is through the company‘ s community forum “So if anyone has a better way of doing something, if they’ve like built a custom tool for like spell checking in your chatbot or something like that, they’ll share it in the forum.
Rasa is a company with great resources and great documentation and tutorials including a dedicated YouTube channel: “We do videos on how to build a chatbot, how to use Raza how to connect it, how to put your chatbot on slack or WhatsApp or telegram or whatever. So we show you exactly how to do that. And that’s all free as well. We also have another YouTube channel that focuses just on the algorithms and the and the machine learning behind Rasa sort of like how everything works under the hood. And that’s free and available as well. But the users are developers, we also have some product teams, because we do have an inner product. We do our product teams, and people working in companies that kind of want to get you know, build the pipeline for their company, and they’re not sure how.”
The value of research in AI
The company also is big proponents of the need to invest in research of AI tools and has a dedicated research team that researchers the latest solutions and the applications of thee to real-world problems. Mady explained: ” So the research goes directly back into our products.” And we’d like to, you know, promote that and encourage it and we also have a research arm in Russia. We have a research team that researches like the latest research, they apply that research to solve real-world problems. So we do that research directly back into the product.”
What happened to conversational AI and where is it at now?
I recalled a time around 2016 where chatbots where the next big thing – I’d received hundreds of press releases and every startup (or conference) seemed heavy into chatbot tech. I asked Mady for her impressions about journey – especially considering AI and conversational AI in particular has a much longer history.
She explained that the company‘s founders started the company during the chatbot craze “because they were trying to build a bot. And they realized that it’s pretty difficult to actually build something that you can take to production and build in a meaningful way. With chatbots in 2016, when you ask a question its ok, but when you asked a followup question it immediately failed. Since then there have been a lot of advances in natural language processing, particularly where context is important in language. When I say something like, ‘the animal crossed the street because it was too wide’ it needs to determine whether width is in reference to the animal or the street.”
Food ordered AI has similar challenges: “If I say, I’m hungry, I want to order some food. Then the chatbot says, Yeah, sure, what are you looking for, or whatever. And then you say, I’m in the mood for sushi. Like it has to maintain that context, right? And I’m just gonna be talking about restaurants and food and whatnot. So we started trying to use natural language processing to solve the problem of context and to solve the problem of negation. For example, you say something like, I would kill to have some sushi right now, the negation is tricky and a chatbot is not going to understand that.”
Thus, Rasa is fpcusing right now on making multi-turn natural-sounding conversations work really well in production, and drive actual business outcomes, for companies that are looking to automate conversations. “We’re trying to focus on getting things to work in production. And moving beyond just proofs of concepts, which is where we were in 2016, 2017. But f you wanted Target or Walmart to have a chatbot, it didn’t work.
So we’re trying to solve that problem of actually going to production. And I think that’s where that’s where the ecosystem is right now. And there are a lot of companies focusing on and on making tools to make some of this process better and easier. So if you’re talking about CI/CD, and some other good engineering practices that apply in conversational AI, there’s lots of companies like GitHub, focusing on making some of that process work better. All of these tools and how the ecosystem got a bit more mature. I think that’s where we’re kind of seeing the industry headed.”
How is conversational AI evolving to include multilingual chatbots?
Rasa has a global community, which is huge in Europe and India. Mady detailed ” In Europe, we have a lot of folks developers trying to create chatbots in German, French and Spanish. In India, we have lots of bots being developed in lots of Indian languages, and not just Hindi or Marathi, but like lots of other languages as well. So Raja framework supports multilingual assistance. And we wanted to do that because our communities are live. We also want to be super-inclusive, to empower citizen developers. Not just European languages in English. So that was a conscious decision that we made. And in our forum, if you check it out, you’ll see that there are lots of developers working on the language models, and they’ll be like, ‘oh, here’s my language model. You know, I’ve also decided to make this open-source, if you want, give it a go.’ And so that has been really helpful because, you know, other people wanting to go chatbots in Hindi can make use of that.
This is also incredibly relevant during COVID-19 where one of Rasa’s users is the World Health Organisation which uses Rasa under the hood. “The WHO chatbot is used by 50 million users. from Arabic to Spanish to Swahili and other languages as well. And so now we’re able to support the development of COVID-19 bots to disseminate information in like a timely way. that’s also accurate. So I think the industry is moving towards making multilingual chatbots. more common.”
If you’d like to hear more from Mady about Rasa, sign up for our upcoming conference – it’s free and promises a fantastic day of speakers and discussion.