One system, taught one module at a time
The end goal is a single AI expert in architecture, construction and the software those trades run on — anywhere in the world. The path there is modular: each jurisdiction's rules, each task, each client it can operate is a module, added where verified data exists and never thrown away. This page is the ledger.
| Knowledge | What it covers | Status |
|---|---|---|
| Indian residential | GHMC Hyderabad building byelaws — margins, coverage, staircases, ventilation | RUNNING |
| Indian statutory corpus | Telangana byelaws · Delhi UBBL · NBC · IS — page-verified facts, each carrying its clause and page locator, feeding the next rule modules | EXTRACTED |
| Vastu | Orientation, placement and zone rules evaluated alongside the byelaws — a first-class constraint, not an afterthought | ENCODED |
| US manufactured homes | California permitting | RUNNING |
| Australia · New South Wales | Pool-safety pre-check and submission preparation; wider residential to follow | BUILT · FINAL CHECKS |
| Client operation | The AI driving professional software directly — Autodesk Revit first, over the Model Context Protocol; Dhruvr Studio already takes plain-language instructions | IN DEVELOPMENT |
Ledger last verified 29 August 2026. Nothing appears here as running until it has been checked against the published source document, and every answer we give shows the clause it came from.
How to read the table
Serving means the module answers through the live engine today. Extracted means the source material has been captured and verified page by page, and is being shaped into rules. Encoded · not yet serving means the rules exist as data but the path that evaluates them is not wired into the served product yet — we say that plainly rather than round it up. Next means the groundwork is done and encoding the rules is what remains.
Where nothing we hold covers your case, you are told exactly that rather than handed a number that looks right. Every gap also tells us what to encode next.
Why modules, and why these
General-purpose AI knows a little about everything and cannot be relied on for the law anywhere. Expertise here is not one big training run. It is thousands of specific, checkable facts — a margin that depends on road width, a barrier height, a ventilation ratio — each either right or wrong. So we train modularly: encode a body of rules, verify it against the source, check it, teach the AI against it, and only then move on.
The first modules are Indian residential, Australian pool safety and Californian manufactured homes for one reason: that is where we hold verified data. Not a market thesis — a data-honesty one. The order of future modules follows the same rule, and the end goal absorbs all of them: one system, any jurisdiction, any task in architecture and construction, operating the software the work actually happens in.
What “client operation” means
Designing a building and driving the software that documents it are usually two different skills. We treat the second as a module of the first: the AI operates the tools — creating the wall in Revit, not just describing it. Dhruvr Studio already accepts a plain-language instruction and produces geometry; the Autodesk Revit integration carries the same idea into the software architects already use, over the Model Context Protocol.
Working somewhere we haven’t encoded yet?
The sequence follows verified data, and verified data starts with practitioners. Tell us where you work and what you would need checked — it directly shapes which module we author next.
Get in touch