What is an Artificial General Engineer?
The public definition of the system category 1UA-AGE is building — and the line between an AGE, a model and an AI agent.
Open reference →1UA-AGE newsroom
Factual updates. Primary sources.
Measured engineering results, stated boundaries and review status from 1UA-AGE in Ukraine.
Permanent reference
Proof and Disclosure Policy
How 1UA-AGE labels evidence, limits public claims, protects implementation details and corrects invalidated statements.
Open reference →Evidence library
Three connected evidence clusters.
Each cluster connects a public definition, a controlled application and the evidence needed to evaluate it. Shared principles remain linked instead of being repeated as unsupported claims.
Artificial General Engineer
The system category, evidence governance and current scope-bounded engineering proof.
Industrial troubleshooting
From unresolved customer constraint to causal model, bounded competitive-edge hypotheses and an engineering validation plan.
Maritime document intelligence
A customer-hosted product and a measurable framework for source-grounded extraction, exceptions and human-controlled acceptance.
Latest factual releaseOdesa, Ukraine
SWE-bench Lite · 68/300 · 22.67%
Building an Artificial General Engineer with Measurable Results
68/300 tasks resolved — 22.67% on SWE-bench Lite, with local Gemma inference and the unmodified official evaluator 4.0.4. Recovery rerun and evaluation conditions disclosed.
Read the measured resultPrevious factual releaseOdesa, Ukraine
AGE landscape · sourced comparison
Capital, evidence and the Artificial General Engineer
A sourced comparison of Prometheus, River AI, SSI and 1UA-AGE that separates direct category comparison, frontier-AI capital, public evidence and our $1B+ strategic valuation thesis.
Read the field note →Previous factual releaseOdesa, Ukraine
Full routing-completion parity · two benchmarks
1UA-AGE reaches full routing-completion parity on two PCB benchmarks
STRF reached 98/98 in three final runs; STM32 reached 211/211 in three final runs. Both match the 100% routing-completion outcome DeepPCB publishes for the corresponding benchmark classes, with explicit evidence boundaries.
Read benchmark result →