Challenge type: classic

Organiser: Tom Mei?ner

1. the initial situation

Modern systems for digitising analogue documents such as OCR (Optical Character Recognition) and VLMs (Vision Language Models) are powerful, but often produce "dirty" data:

• Character errors: "0" instead of "O", "l" instead of "I".
• Loss of structure: Tables are flattened, line breaks are destroyed.
• Context errors: Technical terms are misinterpreted due to poor image quality.

2. the challenge

The challenge is to develop an intelligent workflow that takes extracted raw data from digitised texts (JSON/text) and improves it using LLMs and algorithmic approaches.

This includes:

• Validation: Does the model recognise whether the extraction is plausible? (e.g. checksums for IBANs, comparison with date formats).
• Correction: Automatic fixing of typing errors (e.g. "invoice9" -> "invoice").
• Enrichment & quality: Add context to missing information and convert the text into a perfectly structured format (e.g. clean Markdown or JSON).

Challenge type: conceptual, creative

Organiser: David Schneider

The aim of this challenge is to design user-centred dashboards for students. These are intended to present their own study progress in an understandable and clear way. The focus is on a central dashboard with various views of important key figures, for example exam registrations, grade progression or average grades. The aim is to clarify which visualisations are really helpful and which key figures offer real added value for students.

An important point is the question of whether and how comparative data is useful. This includes, for example, comparing one's own performance with the cohort or the degree programme average. Possible psychological effects must also be considered here. In addition, the benefits of explanatory and supporting elements should be examined, for example through a linked language model. These could help to better understand the data and point to suitable support programmes.

The results of the challenge should help to make students' study progress more transparent and, if necessary, make suitable Central Student Advisory Service tenders more visible.

Challenge type: creative, conceptual

Organiser: Charlotte Steinke

Students and teaching staff can use the university marketplaceExternal link to find tenders in the subject) area|field of digital teaching at the University of Jena. Similar to an online shop, you can find qualification opportunities, exchange and counselling services or digital tools, for example, and narrow down the search using appropriate filters.
Many similar platforms today have AI chatbots that can answer users' questions directly and refer them to suitable tenders.
The challenge aims to develop a concept and an initial prototype for an AI chatbot that facilitates the search for suitable tenders in the university marketplace and responds to individual enquiries.
The aim is to define requirements for the system from the user's perspective and test its implementation with the university's available AI infrastructure.

Challenge type: classic

Organiser: City of Jena

The city collects vast amounts of data daily through sensors – from air quality and traffic density to weather conditions. These data are essential for planning, monitoring, and public services. Yet the enormous data volumes (up to terabytes per year) consume significant storage, slow down analysis, and lead to high costs and substantial energy consumption, a challenge from the perspective of ecological sustainability. In most cases, not all collected data are actually necessary for the specific question at hand.

The Challenge:
Develop approaches that allow for a meaningful reduction of data volume without compromising the usability of the data for important applications. Key guiding questions include:

• Which information is truly relevant for different use cases?
• How can data be summarized (aggregated, compressed) while preserving important trends and patterns?

Practical, reproducible solutions are sought that will help the city of Jena store its data more efficiently, cost-effectively, and sustainably.