Fraunhofer Institute for Experimental Software Engineering IESE“Data Treasure AI”: Using generative AI to tap into untapped data treasures
Project for the automated creation of training content from existing data using generative AI.
Last modified:
Fraunhofer IRB Success Story – Data Treasure AI
Fraunhofer IRB Success Story at a glance
What it's about:
Innovative tool in research project
The challenge:
Context engineering – data preparation for generative AI agents
The support:
Development and evaluation of generative AI applications
The result:
Test-ready prototype
What it's about: Innovative tool in research project
The Data Treasure AI research project has developed a tool that extracts and groups topics from selected PDF data,
- creates precise summaries of relevant topics with references to data sources, and
- extracts relevant facts with page-specific references, minimizing hallucinations.
- It generates question-answer pairs and multiple-choice tasks (e.g., for further training) and
- provides an intelligent search engine (RAG).
High data quality is essential for all applications. The project analyzes unused company data and provides structured information. It addresses questions about the efficient use of data and ensuring knowledge transfer, e.g., during onboarding.
The challenge: Context engineering – data preparation for generative AI agents
The challenges are:
- Extracting information such as topics, facts, and questions with answers (with varying degrees of difficulty) from existing textbooks (PDF).
- Analyzing multimodal data (text, images, tables).
- Making content searchable for an AI agent.
- Making the extracted and generated content available to users in an interactive way.
The support: Development and evaluation of generative AI applications
Design, development, and evaluation of generative AI-based applications. Including data preparation, database, search engine, agents, and graphical user interface.
That's according to Julia Ehl, Fraunhofer Information Center for Planning and Building IRB.
“The collaboration with Fraunhofer IESE was extremely valuable for us. Thanks to their expertise in generative AI, we were able to develop innovative solutions that help us use our data efficiently and transfer knowledge effectively with the tool we developed. Fraunhofer IESE's support has significantly advanced our project and opened up new perspectives for us.”
Julia Ehl, Fraunhofer Information Center for Planning and Building IRB
The result: Test-ready prototype
The prototype developed by IESE is currently being tested with potential users. Fraunhofer IRB has already integrated the results of the project [in German] into its portfolio.