Robots and Neural Networks in Architecture: How Our Meeting at AS&А of KFU Went

On April 7, 2026, at the Academy of Construction and Architecture of the V.I. Vernadsky Crimean Federal University, we presented two of our projects — “AI Architect of Crimea” and “Tavrida Robotics in Residential Construction”. The audience was students of the 3rd and 4th years of the architecture faculty and teachers of the department of information technologies in construction.

What we showed

AI Architect of Crimea

A neural network that automates the initial stages of architectural design: from a brief and terrain map, it generates 5–7 variants of building placement, taking into account norms, insolation, wind load, and Crimean climate specifics. The project was born from a collaboration with the Architectural Heritage Studio — to recreate the historical appearance of buildings that were lost during the war or in Soviet times.

For the demonstration at AS&A we took one of the case studies: a plot in the center of Simferopol where there used to be a merchant’s mansion from the early 20th century. The neural network generated variants of the mansion’s reconstruction based on archival photographs — and the result was interesting: the proportions and the location of windows corresponded to the historical photographs, but the network added elements that the architects themselves would not have thought of (a winter garden on the roof, a courtyard gallery).

Tavrida Robotics in Residential Construction

An applied project for construction sites: a robot for pouring foundations, bricklaying, plastering. The development is at the concept stage, but the architecture students showed genuine interest — many saw this as a future where the construction site is no longer a place of heavy physical labor, but a place of operating robotic systems.

What surprised us

The students had come prepared: they had questions about integration with BIM, about the legal status of architectural decisions made by AI, about the ethical boundaries of using neural networks in design. This is a new level of dialogue — three years ago we would have talked about “what is AI in general”, today we are discussing specific architectural and legal problems.

We were also surprised by the engineering preparation of the students. When we showed the AI Architect working on a real Crimean case, one of the students immediately pointed out a problem with the placement of the building relative to the red lines — the neural network had not taken into account the actual boundaries of the land plot. This is a known limitation of the current version, but the speed with which the future architect found the error was impressive.

What’s next

We agreed with the leadership of AS&A on three joint initiatives:

1. Practice for students at Tavrida Robotics. Starting from September 2026, 3rd-year students will be able to do industrial practice at our company — working on real cases of architectural AI.

2. Joint research. In October 2026, we are launching a study of the applicability of neural networks in the restoration of architectural heritage of Crimea. The database of lost buildings will be formed jointly with the university.

3. Open lectures. Once a quarter we will conduct an open lecture at AS&A on the topic of AI in architecture and robotics in construction. The first one is scheduled for October.

Why is this important

Crimean architecture has its own specifics — the combination of historical heritage, climate, regulatory framework. Generic AI tools (Midjourney, DALL-E, Stable Diffusion) give a general result that does not take into account the Crimean context. We want to build a local model that knows Crimean architecture, history, and norms — and trains on local data.

The students of AS&A are the target audience for this model. By working with them now, we are forming a community of users who understand the technology and can apply it competently in their work.

If you did not attend the presentation, but want to participate — write to us via our contact form or call +7 979 040-69-41. Robots and neural networks do not wait!