NCC Romania (RoNCC) is part of the EuroCC 2 project, which establishes national High Performance Computing (HPC) competence centres across Europe. Its role is to support Romanian business, public-sector and research users in the adoption of HPC, data analytics and artificial intelligence by facilitating access to expertise and computing resources at national and European level. Within this framework, RoNCC has supported a collaboration with the Agentia pentru Dezvoltare Regională București-Ilfov (ADRBI) and the PROSME Enterprise Europe Network (EEN) consortium, focused on improving the drafting of partnership opportunity profiles used by EEN advisors across Europe.
ADRBI is a Romanian regional development organisation and the former coordinator of the PROSME EEN project, a consortium of 14 national and regional organisations delivering Enterprise Europe Network services. One of the key tasks of EEN advisors is the preparation of partnership opportunity profiles for publication in the EEN Partnership Opportunity Database (POD). These profiles must follow a predefined structure and are drafted in English.
In practice, many advisors work under time constraints and are non-native English speakers. This can lead to inconsistent language quality, imprecise terminology and multiple review cycles before a profile is accepted. The objective of the collaboration was to improve the consistency and clarity of these profiles without increasing the drafting effort required from advisors.
With RoNCC support, ADRBI evaluated several open-source Large Language Models (LLMs) as drafting assistance tools. The models are being fine-tuned and/or prompted using a vector database trained on a curated dataset of approximately 1,000 previously accepted partnership profiles from the POD. HPC resources are used for data preprocessing, model fine-tuning and evaluation.
The resulting prototype is a web-based drafting aid that provides real-time text suggestions, terminology checks and alerts for deviations from the required profile structure. A local pilot deployment indicated a reduction in average drafting time from approximately 45 minutes to around 32 minutes, along with improved consistency and fewer revision cycles during internal review.
The solution is designed to be deployed using shared HPC infrastructure and an open-source software stack, limiting operational costs and avoiding licence dependencies. The prototype is technically ready for deployment across the 14 PROSME EEN partner organisations, subject to further validation and rollout decisions.
