AI Will Help Build Faster
The Far East and Arctic Development Corporation has integrated a GigaChat-based AI assistant into its master plan management information system. The assistant reviews milestone reports, analyzes supporting documentation, and can request additional evidence from contractors to confirm that work has been completed.
During the first quarter of 2026, the system operated alongside human reviewers, duplicating their work. Human reviewers found the AI's decisions to be correct in more than 90% of cases. That makes it a notable example of using a Russian-developed generative AI model to verify project documentation and support large-scale infrastructure programs. More than 900 organizations are implementing the master plans, with over 1,800 facilities under construction. The platform is expected to oversee about 100,000 project milestones while processing up to 10,000 reports every quarter. In effect, integrating the GigaChat-based AI assistant into the master plan management system is expected to accelerate the construction of hospitals, roads, utility networks, educational facilities, and urban infrastructure while reducing the administrative workload for human reviewers.
Human Oversight Will Remain
The next objective is to move from pilot use to fully automated acceptance of standard project milestones. Still, disputed, financially significant, and high-risk cases are best left to human specialists. Although an agreement rate of more than 90% represents a strong pilot result, it does not completely eliminate the possibility of incorrect approvals. The next stage could combine text-based reporting with mapping information, photographs, video, and geospatial data.
The system was designed around Russian government workflows and information environments, so its export potential remains uncertain for now. If its economic value is confirmed, individual technologies such as project documentation analysis, schedule monitoring, and automated report verification could eventually be packaged as standalone products. Those capabilities could support the development of what is often described as a "digital government." Even so, the immediate priority is to expand the system to additional Russian cities.
Kamchatka and Sakhalin Become Pilot Regions
The digital environment for managing master plans was established in 2024. Project managers began entering activities from long-term development plans for cities across Russia's Far East into the system. At the same time, work began to integrate the platform with the National Spatial Data System. Kamchatka Krai and Sakhalin Oblast were selected as the first pilot regions.
In 2025, the Russian government introduced AI into the monitoring of national projects and government programs. That same year, the Federal Treasury added a machine learning tool to the GAS Upravleniye (State Automated System "Management") platform that predicts whether national project milestones are at risk of not being completed. According to the Treasury's final report, the model achieved forecasting accuracy of up to 90%.
In 2026, the Federal Treasury also began using a domestic platform to analyze video from construction sites. The AI monitors the number of workers on site, the operations being performed, equipment usage, and potential deviations from project schedules.
Multimodal Oversight
The neural network now does more than summarize information. It also verifies evidence confirming that specific phases of a large-scale government program have been completed. Over the next two to three years, similar tools are likely to be deployed across regional project offices, government construction oversight systems, and territorial development programs. The most promising direction is multimodal oversight, where AI simultaneously compares reports, site photographs, video footage, registry records, and geospatial data.
Ultimately, success will be measured by tangible reductions in project delays, rework, and unjustified public spending. The Far East and Arctic Development Corporation will evaluate not only the number of documents processed but also the number of violations detected, the amount of specialists' time saved, and the effectiveness of the appeal mechanism for AI decisions.
