Odessa-based deeptech project 1UA-AGE has reported the first reproducible results of internal testing of its AI agent for electronics development. The autonomous system achieved 100% connection tracing on two complex benchmarks — a STRF radio frequency circuit and a four-layer STM32 board — in the KiCad environment.
Odessa-based deeptech project 1UA-AGE has reported the first reproducible results of internal testing of its AI agent for electronics development. The autonomous system achieved 100% connection tracing on two complex benchmarks — a STRF radio frequency circuit and a four-layer STM32 board — in the KiCad environment.
During three consecutive runs, the AI engineer laid all collision links without any new violations of the engineering design rules (DRC). This was reported to dev.ua by the founder of the project, Yevgeny Vasyliev. According to him, the developers have prepared a controlled evidence package to be transferred to an independent engineer to verify the results.
The main goal of 1UA-AGE is to create a so-called Artificial General Engineer capable of performing engineering tasks in the physical world with the ability to clearly verify each step. As part of the final stage of internal testing, the model successfully passed two printed circuit board (PCB) benchmarks:
STRF scheme for mixed and RF signals: 98/98 laid connections in each of the three runs.
STM32 4-layer board: 211/211 connections in a fixed KiCad environment with no new design rule violations (DRCs).
The team notes that these are internal synthetic results that have not yet undergone independent physical or hardware verification. The startup founder emphasizes the importance of transparency and provability in engineering AI.
“Models suggest. Engineering tools verify. Evidence preserves what actually happened. Artificial intelligence can generate. Engineering must prove,” notes Yevgeny Vasyliev.
The company also refrains from making direct claims of superiority over other solutions on the market, such as DeepPCB, as testing was conducted in different CAD models and measurement conditions.
The development of AI agents in the field of hardware design is gaining momentum. Recently, OpenAI demonstrated the capabilities of GPT-6 Astra, which is able to lay out a printed circuit board and lay copper connections in KiCad in 3 minutes.
