Smart “Orchestrator”
Neural networks have not yet learned to independently design safe highways, but they are already helping manage the process by taking routine work with regulatory documentation and fragmented software off engineers’ hands.
Engineers no longer need to spend hours cross-checking multivolume GOST standards and manually transferring data from one calculation module to another. Today, they can simply describe a task in ordinary language, and an intelligent assistant will determine which standards apply, which programs to run and how to combine the results. Scientists at the Moscow Automobile and Road Construction State Technical University (MADI) have proposed just such a system for automating engineering procedures.
At the heart of the development is an LLM agent that acts as an “orchestrator.” It does not generate engineering parameters itself. Instead, it coordinates specialized calculation software, checks compliance with applicable standards and explains the reasoning behind each decision. This is a crucial distinction from standard generative models, which can produce plausible answers that are nevertheless physically or normatively incorrect.
The developers emphasize that their system makes every stage of the design process transparent by identifying the basis for decisions, a critical requirement for using AI in safety-sensitive engineering processes. According to the researchers, this approach can make routine design procedures several times faster.
Domestic Evolutionary Path
The development direction closely matches the digital transformation agenda for the road sector. A key advantage of the MADI system is its compatibility with Russian CAD systems and corporate knowledge-management platforms. This opens the way for a new class of domestic engineering software: intelligent layers built on top of existing design systems, calculation suites and regulatory databases. Given that the Federal Road Transport Agency had already moved most of its key information systems to software independent of foreign suppliers by 2023, this approach is particularly timely.
Looking Ahead
In the near term, the developers plan to expand the system’s capabilities by adding specialized models for geotechnical engineering, hydrology and construction management, as well as integrating BIM data. This will allow the system to cover more stages of road project preparation and fit it into a unified digital environment for design organizations.
The technology has export potential, but realizing it will require adaptation to the regulations, language and software used in each individual country. The most natural markets are likely to be the CIS and Eurasian Economic Union countries, where engineering standards remain closely aligned.
Historical Context
Russia’s road sector has been steadily moving toward solutions of this kind. In 2021, the Ministry of Transport and Avtodor began an active transition to digital project management and information modeling, having created 35 information models of construction projects by that time. By 2023, the Federal Road Transport Agency was already using an AI model to analyze the safety of federal highways, processing data from more than 2,000 kilometers of roads. In 2024–2025, the agency also systematized requirements for using BIM and AI and prepared methodological recommendations.
The global context also underscores the relevance of this approach. International research is increasingly exploring ways to link LLMs with BIM for automated building-code compliance checks, with the language model acting as an intermediary between regulatory texts and engineering tools.
Conclusions and Outlook
The MADI software demonstrates a shift from general-purpose chatbots to specialized AI agents embedded in professional engineering systems. Its key advantage is the architecture: calculations remain with verifiable software, while the LLM handles coordination, selection of applicable standards and interpretation of results. That is much closer to the requirements of industrial AI deployment than conventional generative assistants.
The most likely development path over the next few years is an expansion in the number of connected calculation modules, integration with Russian CAD and BIM systems, and pilot deployments at design organizations. If the technology's economic benefits are confirmed on real-world projects, it could become part of larger domestic automated design systems.
