Modellierungs-Fitnessclub is a hands on meeting where researchers build modelling workflows with coding harnesses (such as Claude code, OpenAI Codex, cursor, Open code or Pi). This is a workshop to practice data science skills transferable between scientific domains. We are all learning how agentic programming is disrupting software engineering and our modelling work as a result.

• The 90 minutes workshop chooses and analyses one dataset in 4 phases:

1. Choice Phase 5 minutes. Round of introduction 1 minute per project (with timer). Group decision on the most mature or desirable project among the submitted ones.
2. Specification Phase 15 minutes. On a drawing board or sheet of paper, we specify what we want to achieve during the session.
3. Implementation Phase 1 hour. Implementation with at least one, preferably two or more laptops which load the dataset, tools, and generate software with a coding agent. We push the code to this git repository under a new sub directory corresponding to the day of the event.
4. Wrap-up Phase 10 minutes. What failed, what went well, what can we improve next time.

• Throughout the specification and implementation phase, we can discuss about visualizations such as histogram, scatter plot, line chart, area chart, bar chart, maps, trees, networks and other graphic representations. We can elaborate a data processing pipeline with data structures (such as multidimensional arrays, vectors and tensors), input/output, file storage formats, databases, git version control and issue management, integration tests, CI, security aspects, workflow engines to run models in a dependency graph and about semantics how we name variables and functions.
• The groups goals could be for example:
• Playing with data visualisations on the dataset of the day. Spawning parallel coding agents with personas from different disciplines to enhance the creative process.
• Translating new research questions into back of the envelope calculations. From these questions, we can brainstorm about the implication for model modification or model linking.
• Performing literature research with LLMs that ground their answers in existing publications such as Perplexity. To use context engineering (uploading relevant papers) and prompting techniques to avoid sycophancy (such as grill-me, expert panel, storm review, hostile review).

Participate in the hands on training camp

To participate preferably come with an open dataset already shared on Zenodo or Figshare or a similar open platform. If you cannot share your data, then find a similar dataset from the literature on which we can work on together so that the learning can be transferable to your own data later on.

• Submit your sample dataset in a new issue also called a work item at:
  https://aidaho-edu.uni-hohenheim.de/gitlab/modellierung/modellierungs-fitnessclu/-/work_items
• Make sure your data file is smaller than 10Mb. If the sample data is larger than 10Mb, please post it in a shared repository that can be accessed online, either publicly or through the university network, for example in an idaho repository. Then share that link in the issue.
• Optionally, you can also add modelling code (in the programming language of your choice)
• If you generate code during the meeting's implementation session you are very welcomed to make a merge request to place your generated code directly in the events/ directory.

If the programming language is proprietary and requires a licence, please come with your own laptop or server access so that you can run the model in it's licensed environment.
At the beginning of each meeting we will vote on which project will be analyzed during that me