Customer Story

Personalised recommendations within a quiz app that makes users smarter

An app that makes you smarter? Trivadis supported the AI Start-Up Mindfire achieve exactly that – with an algorithm that recommends content to users which interest them and widen their horizon.

In Short


Provide the users of a quiz app with a personalised user experience.


Establish a recommendation system which is based on previously viewed content and similar user profiles.


The personalised recommendations enable users to acquire new knowledge by challenging them in areas of their interest. At the same time, such recommendations increase the interaction rate with the app.

Our solution

Artificial intelligence that expands the boundaries of human potential – that is the vision of the AI Start-Up Mindfire. In accordance with this vision, Mindfire created the quiz app «Fire42» that aims to make its users smarter. Users can answer insightful and entertaining questions about topics that they have not necessarily known about before. For example: Why do flamingos stand on one leg? Users are also able to upload their own questions to share their knowledge with the community and challenge their friends and followers.

For users to learn something new, it is crucial to display questions to them that correlate with their interest and they have not yet seen. This is what the Trivadis team accomplished for Mindfire.

Widen users’ horizon and simultaneously spark interest

Mindfire was looking for an algorithm that recommends questions to the users based on their interests. The aim was to widen users’ horizon and encourage them to acquire new knowledge. Through personalised recommendations, the app could generate an individual user experience as well as more interactions.

Hence, Trivadis developed a collaborative filtering recommendation model that enables the app to suggest questions from within various topical domains. This recommendation model bases its recommendations on users’ implicit behavior within the app.

Recommendations based on previous engagement and similar user profiles

First, users are assumed to be interested in a question, if they engage with it in any way, for example by answering or liking it. For every user, a personal profile is built based on their past interactions with various questions. In a second step, their profile is compared with those of other users. Further recommendations are made based on the interests of users that have a similar profile and engagement with similar questions in the past.

I am notably delighted to see, the energy that my young and talented team invested into training the custom model and constructing the production pipeline. The system flourished into providing users with an extraordinary opportunity to receive personalized challenges to enhance their knowledge horizon.

Parinaz Ameri, Senior Consultant, Trivadis

Independent software solution and automatic retraining

Trivadis established the entire training machine learning pipeline in order to develop the recommendation algorithm. This pipeline is built by means of open-source software and based on Trivadis’ AI/ML JumpStart architecture. The fact that the architecture solution is based on open-source software enables Mindfire to freely decide if they want to move their training environment from on-premises to a virtual machine in cloud environment. They are not bound to any provider.

The production pipeline, which enables the utilization of the developed recommendation model within the app, is however orchestrated exclusively by means of Google Cloud services. Once the recommendation model is in production, it must be retrained on a daily basis to cover the newly generated data by recent user activities within the app. This retraining process is entirely automated within the established pipeline.

Fully orchestrated pipelines and an autonomous Data Science team

In addition to developing a fully functional recommendation system, Trivadis provided Mindfire with an orchestrated training and production pipeline for their current and future Machine Learning projects. Trivadis also established a Data Science team for Mindfire and coached their members to not only operate the already established pipelines, but also develop new models.

About Mindfire

The Mindfire AG is part of the Mindfire Foundation, whose goal is to make fundamental advances in human-centered AI and expand human potential. Founded in 2019 in Pfäffikon, Switzerland, the startup develops tools and technologies that help achieve this goal. In October 2020, the Mindfire Foundation launched the first Swiss AI Award which recognizes the most promising and innovative AI startups in Switzerland.

Technologies used

  • Google Cloud Firestore
  • Google Analytics
  • Vertex AI
  • Google Cloud Functions
  • Google Cloud Pub/Sub
  • MLflow
  • Optuna
  • DVC
  • Docker-compose
  • Python
  • Jupyter Notebook

More stories

AI Cloud IoT Analytics Data Platform Microsoft Health Care
How artificial intelligence helps children breathe
Data Platform Cloud Automation Microsoft Automotive Analytics
With central data plat­form to the self­driving car
Data Platform IoT Cloud Microsoft Public Administration Analytics
Platform analyses and visualises large amounts of data

Your contact