Recently Tasytt launched Obie: a Slack chatbot for company knowledge. Teams can ask "what", "how", or "where" questions, such as "What is our computer policy?". Obie either finds the answer in one of your documents, or will ask you to provide him with the answer so he can give it next time someone asks the same question.
Obie has integrations with several existing services: Google Docs, Confluence, Google sites, Evernote and Dropbox. This means companies don't have to start from the ground up with training Obie. Giving Obie access to this existing knowledge ensures a short training period.
InfoQ reached out to founder and CEO Chris Buttenham to ask him some questions about Obie.
InfoQ: We tried Obie a little bit in our Slack, but had the idea that it did not analyze our previous conversations. Is this a feature that will be added in a future version?
Christ Buttenham: You could probably imagine you’re not the first to ask this question! Although it would seem that content living in Slack would be a natural place Obie would start, we actually feel that most conversations are fairly unstructured and somewhat useless when it comes to organizing team knowledge. We’re definitely considering adding content that lives within Slack as something Obie can reference, but we feel the low hanging fruit is the rich content scattered across multiple sources within an organization.
InfoQ: Right now Obie only reads from the knowledge base you made yourself. Are there any plans to add extra intelligence like Alexa and Siri have?
Buttenham: We’re fairly focused on the concept of "internal" knowledge. You never know which direction it could head, but our vision is to be the most intelligent solution for learning anything internally. That means we’re interested in much more than just company knowledge; it could be any team, family or individual. Imagine you want to share something in a tacit way with someone you know—we envision that they just ask your personal Obie.
InfoQ: There are many ways people ask the same question over and over in different ways. How does Obie recognize the intent of a question?
Buttenham: We use a combination of third-party and our own in-house NLP (natural language processing) technology in order to understand intent, context, keywords and other metadata gathered via Obie. We also use fallback mechanisms on various services for this.
InfoQ: Natural language processing is a hot topic at the moment. Could you give us a peek under the hood of Obie and tell us what techniques it uses?
Buttenham: Obie uses a combination of techniques to ensure that the most appropriate results are returned. These techniques involve a scoring pipeline where each knowledge entity is assigned a score based on a number of attributes and metadata. To increase the accuracy, Obie continuously learns based on user behaviour with the results. Entity classification is used to categorize documents at a more abstract level and keyword extraction is used in combination with word vectors to ensure alternative results are captured. Lucene-like scoring mechanisms are also used to pre-process results and achieve greater performance with larger knowledge bases.
InfoQ: We found that Obie can integrate with existing "knowledge-bases" such as Dropbox, Onedrive, and Atlassians Confluence. If we as a company have very important sensitive private information in such places, would it still be ok for us to let Obie analyze this data?
Buttenham: Absolutely, and we get that question a lot. We take security and privacy very seriously - strict access control policies are in place to ensure only authorized servers have access to your information. None of the information from the integrations are stored on our servers and we provide the option to encrypt all the content authored in our Dashboard/CMS. The encryption keys are handled securely and rotated frequently on secure infrastructure.
InfoQ: There are still many teams out there who are not using Slack, instead relying on HipChat or even old fashioned email. Are you already thinking about ways to use Obie for other chat clients?
Buttenham: There is definitely a market outside of Slack. With a small team, we can’t boil the ocean so Slack provides a great starting ground for Obie with a highly growing sample set of our ideal customer segment. We partnered with message.io and Microsoft to bring Obie to Microsoft Teams soon, with the ability to easily port to HipChat, Skype, Cisco Spark among others. We’re building Obie so that one day he can be accessed anywhere conversation lives—even SMS!
InfoQ: What gave you the idea to build such a bot?
Buttenham: It was totally by mistake! My co-founders and I first experienced this problem at the companies we were at previously and decided to build an app to solve it. We quickly realized that teams don’t need more tools! The surge of bots and the conversational interface was the lightbulb moment that turned our product into Obie; a chatbot.
InfoQ: You stated that other chatbots are disrupting existing workflow processes. We were wondering if you could give an example of this, and why Obie is better.
Buttenham: If I’m understanding the question correctly, there is a common misconception with bots; that they are an all encompassing solution. We are better than our competitors because of our technology, user experience, and brand, but there are plenty of non-competing bots out there. For example, Growbot (growbot.io) is a bot that helps teams celebrate wins and facilitate "props" or "kudos". There is plenty of room for bots to be point solutions in our opinion so that we can focus on solving one problem, really really well. You can see parallels from the early "app" days to the surge of the bot space.
InfoQ: Imagine that another developer also wants to start a Slack integration service. Are there things you wish you knew before you started?
Buttenham: Just that if you're going to have a complex bot, it's all so new that you're going to be going through some growing pains along with the rest of the industry. Web and mobile apps are well-established, therefore there are set standards. Bots like Obie are fairly new and it's an entirely new way of interacting with software. It’s unlikely you’ll get it all right the first time, so make sure you stay on top of new API features and blogs about trends and development tools emerging in the industry.
Don't let the seeming simplicity of a bot fool you: it's like a webpage with nothing but inputs. A lot of user input is always the toughest to handle, and always prone to users not doing what you expect. For example, you can't just disable a button if you don't want it clicked; your bot has to handle all inputs and do so gracefully to ensure a good user experience.
