<- all articles

Oswald 2019 feature wrap-up

Michiel Vandendriessche

Now that 2019 is coming to a close, we wanted to give a final update about our beloved Oswald chatbot platform. There are so many things to talk about with this release. Let's dig in!

The best Oswald yet ⚡️

Not only did we make Oswald faster and more future-proof than ever, but we also have a truckload of new features to show you.

What we've worked on

  • The hybrid cloud (beta)
  • Response debugging (beta)
  • Twilio: SMS and voice enabled chatbots (beta)
  • Confusion matrix
  • WhatsApp integration (beta)
  • Microsoft Teams integration (beta)
  • Import and export of intents, entities and scenarios
  • Smoke tests

The hybrid cloud, finally 🤯

In some corporations, it is an absolute necessity to keep data and/or APIs from reaching the cloud, whether it is a regulatory restriction or a company policy. Previously, Oswald was required to be installed and run completely on-premise, making it difficult to update the platform and add features which, in today's ever-improving technology landscape, is a substantial drawback.

Our new hybrid cloud allows you to run code responses on-premise. The code response can interact with sensitive APIs without talking to the chatbot platform itself. This is a huge improvement and cures a big pain for some of our customers.

Response debugging 😌

You can now debug your code responses in your favourite IDE instead of the browser, which means you can make use of breakpoints and all that good stuff, which makes life a little easier.

Code response example in the browser

Twilio: SMS and Voice enabled chatbots 🙊

Our brand-new Twilio integration lets you interact with your chatbot via SMS. (You can try it yourself)

IVR on steroids

Twilio Media Streams introduced an even more powerful addition to our platform. You can now call your chatbot, potentially replacing your Interactive Voice Response (IVR) system entirely!

With real-time intent detection in place, the commercial applications are endless. First use-cases that come to mind:

  • Oswald can now be used to provide a caller with a more sophisticated experience than simply an automatic call distributor.
  • Salespeople can input data into their CRM tool by voice during a phone conversation with Oswald, e.g. while driving from one customer to the next.

Another powerful aspect of this integration is that, instead of using the default Twilio speech-to-text solution, we direct the incoming audio stream to whichever provider has the best transcriptions for the spoken language. We initially used this approach to deliver the best experience to our Dutch customers, but as a consequence, a lot of languages that are not widely spoken are supported. As long as there's a suitable enough speech-to-text service for it.

Confusion matrix 🤔

In the NLP section of your chatbot, you can now gain insights into how well your intents are being recognised. 80% of your bot's training sentences are used to train the model and 20% are used to verify whether the correct mappings occur. This makes it much more obvious where more (or less) training data is required.

An example of a good matrix. In a 'bad' chatbot, the diagonal would not be as prominent

WhatsApp integration 📞

Ever wanted to connect your Oswald chatbot to the most commonly used messaging platform in the world? Adding our new Tyntec integration gives your chatbot WhatsApp superpowers!

Simply register your account and phone number and fill out the screen below and you're set.

Integration screen

But what if your chatbot isn't ready for production yet, and you only want your bot to respond to specific phone numbers?

By toggling the "customize-senders" option, you can add a list of phone numbers as testers. The chatbot will only respond to these numbers, and messages from all other phone numbers will be ignored. Neat, right?

Customizing senders

How simple is that?

Microsoft Teams integration 👨‍👩‍👦

Now we all know that Slack's chatbot, Slackbot, can do things based on commands it receives.

Slackbot message

But what if you wanted to ask natural questions like:

  • Who is responsible for human resources?
  • Where can I find our company's brand book?
  • How do I connect to our Dropbox?

In most companies, to find the answers to these questions, you'd have to send approximately 52 emails, 36 direct messages and follow 7 links to eventually find the answer to these questions. If it takes you 30 seconds to open a link, 20 seconds to open every email and 10 seconds to send a direct message, you'd lose a whopping 27 minutes of your precious time.

A possible solution

Oswald already had a Slack integration, but now we've added a Microsoft Teams integration as well, so you can seamlessly integrate your Oswald chatbot into your Teams environment.

Microsoft Teams chatbot

Customer support

It's clear we all want a chatbot for every possible question within our company, but conversational interfaces are creating enormous waves in the world of customer support as well and can greatly reduce the workload of customer support staff.

Imagine having a channel in Teams named "support" and this cute guy pops up!

Oswald takeover card

This card tells us that someone needs help. Clicking "Takeover" will start a direct conversation between you and the user, without any interference of the chatbot. Once the user's question is answered, the chatbot will take over again to answer the user's questions.

So what are you waiting for?

Import and export of intents, entities and scenarios ♻️

In addition to the existing importing and exporting of a bot's entire configuration, we've now added the same feature for specific intents, entities and scenario's.

This allows you to easily copy-paste entire parts of your chatbots from one chatbot to another, or duplicate them by exporting and importing into the same chatbot.

Import and export entities from the entities menu.

Smoke tests 🔥

Save conversations with your chatbot and replay them after some changes to verify whether the responses are still correct.

Smoke test demo

Coming up in 2020

Oswald suits up

Since an interface redesign is long overdue, we'll be rolling out a complete UI redesign starting January. The platform will also be getting the UX it deserves with better error messages and improved tracebacks. Together with the newly introduced response debugging, creators can now focus entirely on the chatbot itself, without having to worry about peripheral matters.

Chatbot versioning

Chatbot versioning will allow you to rollback to any previously published version of your chatbot.

File upload

Because of popular demand, Oswald chatbots will soon be able to accept uploaded files.

Improved search

Your entire chatbot will soon be indexed by Elastic Search so you can search through intents, entities and all stuff related in a single search.

Shared libraries

Want to reuse code snippets across one or more chatbots? Just save them to and retrieve them from shared libraries!

Getting started

Hyped? Want to take a look at this amazing chatbot platform?
Not to worry! You can reach us via hello@oswald.ai for a demo or send us a tweet.

Written by

Michiel Vandendriessche

Want to know more?

Related articles