MAKEathon 2024: Innovative Problem-Solving with AI, FHNW School of Business

The fourth edition of the MAKEathon event, held at the FHNW on October 26th and 27th and led by Dr. Emanuele Laurenzi from the Intelligent Information Systems research group brought together over 50 students who collaborated to tackle eight challenges presented by various companies and with the support of more than 10 AI experts from both research and industry. This dynamic two-day hackathon showcased innovative problem-solving across diverse industries, focusing on hybrid Artificial Intelligence, i.e., combining data-driven AI and knowledge-driven AI methods.

In total 11 innovative AI prototypes were presented on the stage of the MAKEathon. Four of the best solutions are summarized below.

• The winning group, Prosciutti Stagionati (comprising master Dual Degree students from Unicam-FHNW: Angelica Berdini, Luca Bianchi, Stanislav Teghipco and Piero Salmena), addressed PostFinance's Challenge 1. Their solution integrated machine learning with SHAP values for interpretability and employed an ontology to strengthen feature relationships. Leveraging Azure OpenAI, the team transformed complex model insights into clear, readable explanations. A custom-built frontend showcased these insights in an accessible format, effectively bridging the gap between advanced AI models and business understanding.
• The second winning group, Turny (formed by BSc Business Artificial Intelligence students Patrick Nydegger, Oliver Gwerder, and Robin Haag), tackled SwissPort's Challenge 7. They developed an hybrid-AI system designed to optimize airport turnaround processes by combining computer vision and process mapping. It leverages Microsoft Azure-trained image recognition models to identify key elements and objects involved in these processes, using BPMN 2.0 to map dependencies and detect delays. A semantic database links essential relationships, such as team activities, while a relational database manages information on aircraft, gates, vehicles, and flights. The system provides a user-friendly graphical interface featuring process overviews, live camera feeds, and real-time alerts with suggested solutions for detected anomalies. Turny supports, rather than replaces, turnaround managers by enhancing early problem detection and improving planning efficiency.
• The group Let’s Fly (master’s students Kyrylo Buga, Fabio Michele De Vitis, Alessio Gesuelli, and Lorenzo Verducci) took on Swiss Airline’s Challenge 3: Drone Aircraft Inspection. They developed a prototype integrating a multimodal agentic chatbot with computer vision capabilities. Their AI model was trained to detect and classify mock-up airplane components, including engines, wings, and tails. Key features of the prototype included: A multimodal chatbot powered by computer vision models; Image segmentation to identify backgrounds, objects, and people;Extraction of specific objects from images captured by a smartphone camera. The prototype was deployed during the event, allowing the audience to test it in real-time using airplane images and the mock-up.
• The group SOBellino (featuring master’s students from the MSc Business Information Systems, Eveline Walker and Nadia Jakob, BSc Business Information Technology student Cliff Clemente, and external PhD student Dana Malcova) addressed SOB’s Challenge 6. They created a low-code solution for a language-learning chatbot aimed at helping German-speaking train drivers learn Italian for better communication with dispatchers in Ticino. The system included a knowledge graph of educational content built in Metaphactory, a conversational flow developed in VoiceFlow, and a Python-based web service for seamless integration between these technologies.
  

The MAKEathon exemplified the power of collaboration and innovation in addressing real-world challenges, with participants showcasing innovative AI solutions. Below is a summary of each challenge

The MAKEathon exemplified the power of collaboration and innovation in addressing real-world challenges, with participants showcasing innovative AI solutions. Below is a summary of each challenge:

Host: PostFinance
Objective: Develop an LLM-powered tool to enhance model explainability for business users. The solution integrates Shapley values, model documentation, and generative AI to generate intuitive explanations in German or English, tailored to applications like fraud prevention and customer personalization.

Host: Metaphacts
Objective: Create an AI assistant to simplify the reproduction of scientific research. Leveraging tools like SemOpenAlex, the assistant should retrieve and analyze software, datasets, and experiments from academic papers. It aims to guide researchers in setup, usage, and running experiments, potentially automating key processes.

Host: SWISS Airlines
Objective: Design a drone-based aircraft inspection pipeline capable of detecting and classifying damages (e.g., lightning strikes, hail blows) from images. The solution must ensure accuracy, speed, and safety while generating detailed inspection reports and recommendations for action.

Host: SWISS Airlines
Objective: Develop a voice assistant to assist customers in real-time during web interactions like booking or resolving issues. The system combines speech-to-text and text-to-speech functionalities and should be adaptable to different web platforms.

Host: IIS Research Group
Objective: Enhance the naturalness of interactions with voicebots by allowing them to understand and respond to interruptions gracefully. Targeted at educational games, the bots simulate stakeholder interviews with the ability to dynamically adjust conversation flow based on user behavior.

Host: SOB (Schweizerische Südostbahn AG)
Objective: Create a low-code platform enabling domain experts, such as language teachers, to develop educational chatbots without technical expertise. The focus is on facilitating language learning for train drivers using AI and domain-specific teaching material.

Host: Swissport
Objective: Build a human-machine interface for a reinforcement learning agent to assist airport dispatchers. The AI system provides real-time monitoring, predictions, and actionable recommendations, fostering a collaborative learning environment between humans and AI.

Host: Baloise Insurance
Objective: Utilize large language models to streamline underwriting workflows. The system should automatically extract insights from client applications, claims data, and external sources, reducing manual effort and improving decision-making accuracy.