Abstract

Several AI-based tools based on large language models (LLMs) have been developed in the last 2-3 years and have started to be used in pre-award to accelerate the proposal writing process. Despite the undoubtful extra help, being dependent on such big commercial players raises questions about the uncertainty of releasing confidential data and knowledge and of potentially getting unreliable pieces of information.
This need for safe and trustworthy AI-based tools has pushed our pre-award and data science groups in the research support team at the University of Southern Denmark Research and Innovation Organization (SDU RIO) to co-design and co-create new solutions for proposal writing.
Such tools are based on locally used open-source LLMs and offer privacy and security within our university. By tuning LLMs according to our specific needs we can create task-optimised assistants.
These assistants can solve less complex but time-consuming tasks, providing us more time for the important steps of proposal writing to rightfully convey research and development concepts and ideas.
In this presentation, we will focus on one of the most important aspects to consider when writing a proposal, that is the coherent narrative flowing along a red thread across all the main sections. Therefore, we will talk about one of the first tools that have been developed: a “coherence check” assistant that can be safely fed with our drafts and then generates a response on the consistency across the whole text.
We will show how we have tested our coherence check on several coordinator Horizon Europe pillar 2 proposals and how we have benefited from it.
Finally, we will conclude with an overview on other task-optimised assistants and with our learning outcomes on their positive impacts on proposal writing.