A Regulatory Code Assistant
for Construction
Construction compliance is fragmented, expensive, and high-stakes. Aspexilary AI is trained on the actual regulatory documents, deployed on your network, and auditable from day one. Not a chatbot — a sealed appliance.
The problem
An airport fuel station engineer needs to know bonding and grounding requirements during aircraft fueling. The answer spans NFPA 407, FAA AC 150/5230-4C, EPA 40 CFR 112, and IATA fuel handling procedures. That's four documents from three agencies — and a senior engineer's afternoon.
This is the reality of construction compliance:
Generic AI hallucinates regulatory answers. ChatGPT, Claude, and Gemini were not trained on the actual code documents. They generate plausible-sounding text that may or may not reflect what NFPA 407 actually requires. For construction, "probably right" is the same as wrong.
What we built
Each Aspexilary domain is a self-contained AI system trained on the actual regulatory documents that govern a specific type of construction. Not web scrapes. Not summaries. The PDFs from FAA, NFPA, EPA, ICAO, IBC, OSHA, and every other relevant agency — extracted, chunked, embedded, and indexed.
Every answer cites the specific section of the specific code. Every query is logged in a hash-chained audit trail. Every container runs on an isolated network with no internet egress — proven by continuous packet capture, not policy assertion.
Why on-premises
Construction firms handle sensitive data: bid documents, structural calculations, client facility layouts, security plans for government buildings. A hospital construction project involves HIPAA-adjacent information. An airport project involves TSA security requirements.
These firms will not send that context to a cloud API. Their legal teams won't allow it. Their government contracts prohibit it.
Our architecture — Docker stack, no internet egress, hash-chained audit logs, tcpdump egress proofs — is not a feature. It's the admission ticket to the regulated construction market.
Five things that set this apart
- Trained on the codes — actual regulatory documents, not web scrapes or training data summaries
- Cites sources — every answer links to the specific section of the specific code
- Runs on your network — Docker Compose, sub-60-second deploy, zero cloud dependency
- Proves isolation — compliance artifacts ship with every domain, auditor-ready from day one
- Scales by domain — one sealed stack per construction type, independently deployable
Your knowledge, built in
Every domain ships with a regulatory baseline — the codes, standards, and advisory circulars that govern that construction type. But regulations are only half the picture. Your firm has its own specifications, lessons learned, internal standards, and project-specific requirements that determine how you actually build.
Aspexilary domains are designed to absorb your proprietary knowledge on top of the regulatory foundation:
- Company specifications — your standard details, preferred materials, vendor requirements, and construction methods get ingested alongside the codes. When an engineer asks about waterproofing, the answer references both the IBC requirement and your firm's approved membrane system.
- Project-specific documents — RFIs, submittal logs, design narratives, owner requirements. Each project can have its own document set layered into the domain, so queries return answers grounded in both the code and the contract.
- Lessons learned — field reports, NCR histories, inspection findings. The knowledge that lives in senior engineers' heads and filing cabinets becomes retrievable by the entire team.
- Internal standards — QA/QC procedures, safety protocols, environmental management plans. Your processes become part of the AI's knowledge base, not just the regulatory minimum.
The ingestion pipeline runs on your side. Documents go in, vectors come out, the knowledge base grows. The regulatory baseline never changes — your layer sits on top. When a code updates, we ship a new baseline. Your custom knowledge persists.
This is the difference between a reference tool and a working tool. A reference tool tells you what the code says. A working tool tells you what the code says, what your firm does about it, and what happened last time on a similar project.
Who buys this
Why nobody else is building this
The intersection of domain-specific fine-tuned LLMs, actual regulatory document corpora, on-premises deployment, auditable compliance infrastructure, and construction industry focus is an empty market.
The big AI companies are chasing enterprise SaaS. The construction tech companies — Procore, Autodesk, Bluebeam — are adding "AI features" to existing products, not building domain-specific retrieval systems from regulatory source documents.
We're building the thing that a $50B construction firm's compliance team will evaluate and say: "This is the first AI vendor that actually read the codes."