Henrik Wiegand

In partnership with BASF and KIT’s Institute for Operations Research, we developed an interactive multimodal dashboard integrating large language models and a chatbot for supply chain data analysis. Leveraging Python, Pandas, and Plotly, the platform processes relational datasets, empowering users to generate real‑time visualizations via natural language queries. A data‑frame agent bridges GPT-4 and the dashboard, translating conversational prompts into executable code and interactive charts. Error‑handling loops ensure reliability, while a tabbed interface offers both graphical and raw data views. Scalable architecture supports future enhancements. This tool exemplifies how to democratize insight generation and lower barriers to data‑driven decision‑making across large organizations.

In collaboration with Fischertechnik and the Institute of Information Management in Engineering (IMI), our team developed an innovative virtual construction kit enabling users to digitally recreate existing Fischertechnik models and design entirely new configurations. By combining CAD-to-VR conversion, a dynamic snapping engine, and an intuitive web-based interface, the platform streamlines 3D assembly workflows. Users can build accurate digital twins of their assemblies before ordering components, reducing errors and enhancing engagement. Integrated within PolyVR, the toolkit supports four snapping types and interactive part rotation. This immersive solution lowers barriers to entry, enriches learning, and elevates customer interaction within the Fischertechnik ecosystem.

This project analysed an automotive paint‑shop colour‑sorting buffer by building a discrete‑event simulation. Following the ASIM methodology we decomposed the system into queues, conveyor cross‑shuttles and parallel storage lines, modelled stochastic machine failures, and implemented the model in AnyLogic. Five controllable levers - line count, buffer capacity, fill level, colour distribution and assignment logic - were parameterised for experimentation. The control merged dynamic scoring with a shuffle heuristic to approximate optimal sequencing. Thirty‑day horizons with multiple replications yielded robust performance maps, feeding a documented decision framework. Stakeholders can now evaluate layout alternatives, buffering policies and scheduling heuristics before costly shop‑floor trials.

As part of a team project at the Institute for Information and Market Engineering, my colleagues and I examined how tailoring chatbots affects user satisfaction. Our study focused on customising chatbot behaviour to users’ neuropsychological profiles—specifically, their preference versus aversion to social interaction and their tendency toward rational versus intuitive decision‑making. The objective was to move beyond the prevailing “one‑size‑fits‑all” paradigm and improve the frequently criticised quality of chatbot interactions. The findings provided the empirical basis for a subsequent publication presented at the 31st European Conference on Information Systems (ECIS).