Generative AI Comes to the Factory
Researchers at the HSE Institute of Artificial Intelligence and Digital Sciences have developed CAD2TechSpec, a system that automatically turns 3D models of parts into finished process plans – step-by-step instructions for machine tools.
The technology is designed to cut the time required to prepare technical documentation in mechanical engineering, aerospace, and other high-tech industries. The research findings were published in PeerJ Computer Science.
Generative AI has long since moved beyond creating images and generating text. Russian researchers have developed a system that turns a three-dimensional model of a part into a complete manufacturing process plan. This is more than a research demonstration: it could be a path to a revolution in manufacturing process planning.
Today, many complex components, including turbine blades, are designed in specialized 3D modeling software. But a model cannot simply be sent to the factory floor. Engineers still have to manually develop a machining sequence that takes into account the part’s geometry, available equipment, tooling, and manufacturing standards. The process is time-consuming and requires substantial expertise.
CAD2TechSpec approaches the problem by taking a three-dimensional model as its input and converting it into a structured description of the manufacturing process that can be read by a computer numerical control (CNC) machine. These machines can use computer instructions to produce parts with extremely complex geometries.
From 3D Model to Machine: How It Works
The core innovation is that CAD2TechSpec “views” a 3D model from 28 angles, reads a text-based assignment, and uses a multimodal language model to map out a machining sequence. To avoid AI hallucinations and produce technically sound plans, the system uses RAG, or Retrieval-Augmented Generation. The approach connects a large language model to external knowledge bases so it can draw on accurate, up-to-date information. The neural network consults ISO standards, equipment data, and a library of validated manufacturing solutions.
Tests on 1,236 models produced an impressive result: mechanical-engineering experts rated the AI-generated plans at 86 to 89 points out of 100, with Mistral Large 3 performing best. The developers are careful to stress one limitation, however: CAD2TechSpec is not an autonomous robot but an AI copilot. It handles computationally intensive routine work, while the final review and responsibility remain with the manufacturing engineer.
Russia’s Market: Enormous Potential
For Russian industry, the technology arrives at a particularly important moment. According to HSE’s Institute for Statistical Studies and Economics of Knowledge, 93% of Russian mechanical-engineering companies already use digital technologies such as CNC machines and computer-aided design (CAD) systems. AI adoption, by contrast, remains limited, ranging from just 2.9% to 14.1%. The gap between the digital hardware already on factory floors and the intelligence layered on top of it leaves substantial room for systems like CAD2TechSpec.
For the public, the benefits would be indirect but tangible: reducing engineering workloads and design times should ultimately lower the cost of finished products. For the government, the technology offers a direct route toward industrial digitalization, particularly with the new PNST 955-2024 (Russian Federation Preliminary National Standard “Artificial Intelligence in Mechanical Engineering. Use Cases”), which has set a framework for AI applications in mechanical engineering since 2025.
Global Context and Export Potential
The global manufacturing industry is moving in the same direction. Tech giants such as Siemens and NVIDIA are actively developing industrial AI operating systems that bring together digital twins and generative design across the product life cycle. Russia is responding to the trend by making AI modules a priority when selecting strategically important industrial projects.
In the medium term, CAD2TechSpec could have export potential. The opportunity would not be to export a standalone neural network, but ready-to-use modules integrated into Russian engineering platforms such as KOMPAS-3D, a Russian computer-aided design (CAD) system for 3D modeling, drafting, and engineering documentation. BRICS countries could become key customers, as Russian engineering software is already being actively promoted in those markets.
From the Lab to the Production Line
For now, CAD2TechSpec is a strong research and technology result rather than a commercial product ready for deployment. The most difficult stage lies ahead: industrial validation at real factories, each with its own mix of machine tools and materials.
Even so, the direction is clear. Neural networks are moving beyond applications that sit around manufacturing and are beginning to work directly with engineering information. Integrating generative AI with Russian CAD systems could create a distinctive domestic software stack capable of competing in the global market. The era when machines were controlled exclusively by people is giving way to a model in which engineers and AI work together.
