MN Bot Makers July Meetup: You Had Me at Hello, World!

Eli Krumholz explains how to implement into a bot’s programming.

One of the biggest challenges with chatbots is creating the illusion of personality and convincing the user that they’re having a meaningful conversation while providing useful information is key to creating a successful chatbot experience. The MN Bot Makers July Meetup tackled that problem with some of the industry’s latest technologies.

Th group got together at Intertech last month to demonstrate tools for incorporating natural language processing (NLP) into their bots. When building a bot from scratch, it can be difficult to plan for every iteration of a question. The question “what are your hours?” is different than “what time do you close?” Without a strong framework, these two questions (and every variation of them) would need to be coded individually for a retail chatbot.

Krumholz explains this way: “ is a very user friendly tool for adding NLP to a chatbot, but its real power shines when it is combined with other services. For instance, making a query to a weather API isn’t hard, but coding all the possible ways a user might ask about the weather would be a pain–that’s where NLP services like are essential.”

When a free framework like is introduced into the code of a bot, the bot can now more intelligently process the questions it’s being asked. Because “what are your hours?” and “what time do you close?” have terms related to time, the bot can respond with a pre-coded statement about hours of operation. It’s not a perfect solution right out of the box, however.

Someone might ask this retail bot, “how much time do people usually spend at your store?” Without training, the bot might respond with hours of operation, detecting the keyword “time.” This can be prevented by telling a bot when it has answered questions correctly. has an interface for training, allowing developers to help it differentiate between meanings when faced with questions–both questions fed from the developer specifically for training purposes and real questions from users.


One of the most useful features Krumholz introduced was’s Prebuilt Agents. They contain a variety of features that can easily integrate into any bot. The Small Talk agent, for example, has answers to prompts including, “hello,” “thank you,” and “how are you?” It’s easy to import, bringing a deeper sense of realism into a chatbot.

To demonstrate the practical applications of NLP, Krumholz has been working with Dave Mathias, a local analytics expert and an organizer for FARCON, the annual financial and retail conference run by MinneAnalytics. They created a Slack bot for the conference, training it with information on presenters, session times, and other frequently asked questions. Mathias estimated it took about a day to enter the relevant information into, training it with multiple versions of question formats to help it intelligently answer whatever question that comes its way.

Creating realistic and useful chat experiences for customers and clients is quickly becoming a necessity. Some even consider chat experiences to be the new web browsers. No matter the application, Minnesota bot developers are taking advantage of state-of-the-industry technology to help create robust, meaningful and useful experiences for users.