A verifi ed knowledge base as a foundation for trust in an expert Artifi cial Intelligence service: the SOVNET and Prooftech IT case

The article examines how the quality and level of verification of a professional knowledge base on the reliability of responses generated by expert systems based on artificial intelligence technologies. Using project management as a case study, the authors test the hypothesis that a specialized system working with professionally verified sources provides higher-quality responses than general-purpose large language models relying on knowledge acquired during pre-training. To test the hypothesis, a prototype of the SOVNET–Prooftech IT service was developed, materials from the scientific journal «Project and Program Management» and the IPMA ICB4 standard. An experimental dataset of 130 open-ended questions covering the competence elements of ICB4 was created, including 74 questions on risk management. A total of 520 responses generated by four AI systems were evaluated by certified SOVNET assessors using blind independent expert assessment. Inter-rater agreement was measured at ICC(2,1) = 0.801. The specialized system achieved a mean score of 9.09 out of 10, outperforming Gemini NotebookLM (7.85), GigaChat (6.07), and DeepSeek (5.56). In addition, existing Russian and international services for searching and analyzing professional publications were reviewed, and an Osterwalder Business Model Canvas was developed to plan the further development and scaling of the solution. The results confirm the value of a specialized, verified knowledge base as a factor in improving the quality of professional AI-generated responses.