Domain-specific AI for regulatory and design questions, grounded in sources you can check, not a general chatbot guessing at UK building law.
Domain-specific AI for AECO
General-purpose AI is not yet a reliable source of truth. Architectural Intelligence combines domain-specific reasoning, adaptive retrieval from primary sources, and independent quality assurance so every answer is evidence-grounded and verifiable.
Adaptive retrieval intelligence
Every query is classified into a retrieval strategy before execution. Broad conceptual questions draw from trained knowledge without forced document search. Specific clause lookups trigger targeted retrieval from the regulatory library. Project-specific questions search uploaded evidence first, then cross-check against statute. The system adapts its retrieval depth to the question, not the other way around.
MECE compliance analysis
The Compliance specialist uses Mutually Exclusive, Collectively Exhaustive analysis to structure regulatory responses. Requirements are broken into distinct, non-overlapping categories, each addressed independently before cross-referencing interactions between them. A completeness check verifies that all relevant Approved Documents have been considered, not just the primary one.
CDM Hierarchy of Control
The Risk specialist follows the CDM 2015 Hierarchy of Control for every hazard assessment: identify the specific hazard, trace the root cause, map all affected parties, then work through eliminate, reduce, isolate, control, and PPE in strict priority order. Residual risk is stated explicitly after controls are applied.
Structured cost comparison
The Cost specialist applies a structured comparison framework with explicit scope definition, like-for-like basis, and breakdown by capital cost, lifecycle cost, and programme impact. Sensitivity analysis identifies which variable has the biggest effect on the comparison, and recommendations state the conditions under which they would change.
Staged design protocol
The Design specialist follows a five-stage protocol for 3D model creation: brief interpretation, staged plan, bounded tool execution, review, and user-confirmed continuation. Each stage produces a visible summary of what changed, what was assumed, and what the next decision point is, so the design process is inspectable and recoverable.
Graduated verification
Every specialist treats user-supplied values as inputs to assess, not as automatically correct facts. A graduated severity ladder separates ordinary assumptions from material concerns, reserving highlighted warnings for genuine safety, compliance, or major factual issues. The system distinguishes between user-provided data, agent inference, and tool-verified evidence.
Multi-pass quality assurance
Cross-domain questions trigger a multi-agent council where specialists work in parallel and an independent Verifier checks outputs for safety, accuracy, and regulatory consistency. Citation links are generated deterministically from source documents, and every significant AI claim can be cross-examined against the cited source directly from the response.