MatPrat: Bot development to keep up with the competition

About the project

MatPrat promotes Norwegian farmers and their products. A big part of their mission is to help schools educate their pupils about cooking and related topics through their three websites.

Recently, they saw a need to keep up with the technological development and decided to develop a chatbot and a voicebot.

The collaboration

NoA Ignite has worked with MatPrat since 2014. In addition to regular maintenance and constant improvements, there are regular campaigns and side projects that require extra resources under shorter periods of time.

Challenge: Rapid technological development

Mats the chatbot was developed for Facebook’s Messenger service. He is becoming better and better at answering food-related questions like “What’s the menu of the week?” and “How do I cook a proper steak?”. It is a highly customised service based on RASA core Natural Language Processing with a constantly evolving set of food-related intents (the purpose behind the users’ input) and entities (the terms that provide the specific context for the intent).

The voicebot is more of an out-of-the-box solution. Based on Dialogflow and our Fulfillment API, it is integrated with Google Assistant and Google Home Hub – if you don’t have Google’s Home device, you can try it out on your smartphone (see instructions under Result below). Ask the bot food-related questions like what you can make for dinner based on the ingredients you have in the fridge.

Google currently invests heavily in voice-based services like Home. New features are added regularly, and they often affect our bots. To keep a good position in Google search results, it is crucial to stay up to date with the technical development. For example, the voicebot was intended to work only on Google Home. However when Google launched Home Hub and its smart screen display, we were challenged to make it look nice.

Another challenge is that bots are far from perfect from the start, and they are not going to learn without the right input. They need to be looked after properly in order to give value to users.

Our approach: Constant improvements

The team in Kraków work actively to keep up to date with current trends and developments, so that we can suggest new and relevant improvements.

It’s important to note that bots are far from perfect in their early stages. In order to improve user experience, it is necessary to analyse traffic and listen to user feedback. Based on the findings, new intents and entities are categorised and created. What are the queries that the bot can’t answer, and what can be done to address those queries?

MatPrat’s editors check regularly which queries the bot fails to answer. Early on, they realised that lots of kids were curious about how to cook a lion and other exotic animals. The editors then created a response to address such questions.

Simple text responses can be set up by the editors themselves, but in more complex cases, they request the code needed from the developers in Kraków. For examples, as many people asked Mats what to cook on a hangover, they decided to give users a selection of suitable recipes that might be appreciated when feeling rough.

Result

The chatbot and voicebot are still fresh out of the oven, but the initial reception has been very positive. Thanks to continuous improvements, the two bots are getting more responsive day by day.

The AI algorithms are very interesting from a technical point of view with heaps of potential for the future.

If you want to test them for yourself, equip yourself with a couple of basic Norwegian food terms or phrases. Mats the chatbot is available through Messenger at https://www.facebook.com/MatPrat/. For the voicebot, simply say “Snakke med MatPrat” in the Google Assistant App on your smartphone.