Smart use of AI tools to inform entrepreneurs: RRA of Šibenik-Knin County with the support of EDIH Adria

Challenges

The Regional Development Agency of Šibenik-Knin County has been acting for years as a key support institution for entrepreneurs and the public sector in the region, providing advisory services, assistance in the preparation of projects and information on tenders and funds.

However, due to the high number of requests for services such as project application or tender assistance, although often repetitive and similar, RRA staff face a burden that hinders personalised access to users. Information on ongoing tenders, deadlines, eligibility criteria, the level of co-financing and relevant documentation must be available quickly and precisely, but the administrative capacity of the agency is limited.

There was a clear need for a solution that would allow faster, automated and scalable distribution of project proposals preparation, without loss of accuracy or quality of support.

Solutions

Through the program Test Before Invest and support EDIH Adria, RRA ŠKŽ tested the application AI Assistant based on large language models (LLM), with the aim of automating the preparation of project proposals (documentation) and thus improving communication with users.

In the test phase, it was developed assistant prototype for the preparation of documentation that:

• Communicating in natural language and recognizing user questions
• Finds and displays relevant information on the preparation of tender documentation from the RRA Šibenik knowledge base
• Automates documentation preparation:
  • Finds previous examples of successful applications for RRA knowledge base competitions used as a starting point for text generation’
  • It has the ability to improve the text that can be interactively entered based on the template for individual competitions
  • It has the ability to communicate with external verification systems such as current prices
• Provides reliable responses based on official documentation and up-to-date sources, eliminating imprecise responses or “hallucinations”

The system uses Retrieval-Augmented Generation (RAG) access: combines the retrieval of information from the database (pdf, web, excel, word) and generative responses of AI models, while preserving accuracy and credibility.

Results and Benefits

Testing an AI assistant confirmed a number of concrete benefits for the agency and users if the production version were to be realized:

• Significant relief for employees – fewer telephone and e-mail queries related to basic information and finding information from the project archives
• Faster availability of information to end-users – in a large number of examples, project documentation can be prepared very quickly, without major delays
• Standardised approach to drafting tender documents – reduced possibility of different interpretations among staff
• 24/7 availability of information from RAG database – users can access information outside working hours
• Possibility of multilingual support – the solution is ready to be disseminated into English, which is particularly important in the context of EU applications and for potential foreign users

The system was tested on the basis of real user queries and showed a high level of accuracy, especially where documentation and tender criteria were publicly available.