Glavgosexpertiza Begins Using Generative AI Tools
Generative AI is moving into specialized government workflows, drawing on industry data and regulatory documents while running models locally.
Russia’s Glavgosexpertiza, the federal institution responsible for state reviews of major construction projects, has begun using a suite of generative artificial intelligence tools in its experts’ day-to-day work, including locally deployed large language models, or LLMs; a retrieval-augmented generation, or RAG, system for searching and analyzing information in corporate documents; a transcription service; and an AI tool for analyzing technical reports. By August, 366 employees across 43 departments were already using the services. Of 6,000 total requests, 61% went to the RAG system and 39% to local LLMs. That distinction matters: The RAG system first searches Glavgosexpertiza’s corporate databases and documents for relevant information and then passes that context to the language model. A separate LLM is already being used to analyze technical reports and identify indications that facilities have undergone reconstruction.
Today, Glavgosexpertiza serves as a testing ground for AI tools. The most successful could later be rolled out to other organizations across Russia’s construction sector.
AI to Review Documents
Several areas already stand out as promising candidates for broader use of AI tools. They include intelligent search across regulatory, technical and project documentation; automated preliminary document reviews to flag potential discrepancies before the main expert assessment; analysis of technical reports, expert conclusions, explanatory notes and other unstructured information; drafting expert comments; and searching the accumulated knowledge base for similar cases. Generative AI could also be connected to the Yedinaya tsifrovaya platforma ekspertizy (Unified Digital Expert Review Platform), GIS EGRZ (Unified State Register of Expert Review Conclusions information system), BIM models and machine-readable project documentation. A suitable database is already being assembled for this work.
Russian developers have strong prospects for capturing the market for specialized models serving industry and government. The methodology and software architecture could also become an attractive option for other countries that need locally deployed AI systems without sending technical documentation to external cloud services.
From Experiments to an AI Engineering Center
Glavgosexpertiza’s current adoption of AI tools builds on years of technology development and experimentation. Back in 2021, the Ekspertiza budushchego 4.0 (Expert Review of the Future 4.0) educational project explored AI-based decision-support systems for reviewing project documentation and information models. In 2023, Glavgosexpertiza introduced semi-automated work with its Baza tipovykh zamechaniy (Database of Standard Comments): In the first quarter alone, it was used to prepare nearly 20% of local expert conclusions. By the end of that year, the organization announced that an AI module had entered production use. Its primary purpose was to automate routine checks.
In 2025, AI tools were piloted in reviews of documentation for three facilities. According to experts, neural networks performed well in speeding up document work and information retrieval. In April 2026, Glavgosexpertiza announced the creation of the Tsentr inzhenerii dannykh i tekhnologiy iskusstvennogo intellekta (Center for Data Engineering and Artificial Intelligence Technologies). It opened June 1, and by August, hundreds of Glavgosexpertiza employees were already using RAG and local LLMs.
Experts Are Still Essential
RAG combined with local LLMs offers an architecture well suited to government agencies and large enterprises: Sensitive information remains within a controlled environment, while the model can work with specialized and up-to-date corporate data. The Glavgosexpertiza case therefore offers an example of how generative AI can be used in Russian organizations that handle large volumes of technical, regulatory and confidential documentation.
Still, AI does not replace the expert. Glavgosexpertiza has consistently stressed that the final professional judgment, as well as verification of the algorithm’s output, must remain with a human. Automation instead helps experts do their work faster, more accurately and with less friction.
