Technology — Vertallax

01 — Platform Architecture

From Raw Input
to Decisive Action

All data — structured, semi-structured, unstructured, external — flows through a single import spine into a canonical operating model that surfaces awareness, enables decisions, and captures execution. The loop closes through continuous learning.

01b — The Principles, in Steel

Six Intrinsic Properties
Built In, Not Bolted On

The sections below explain the architecture in detail. These are the principles the substrate is built on — each one intrinsic to the foundation: a structural commitment, not a feature toggle.

Orthogonal

Independent dimensions. Clean intersections.

Every dimension of the model — time, structure, money, judgment — varies independently and intersects cleanly. No field doing double duty, no meaning smuggled into a naming convention. When the dimensions are orthogonal, every question has an address.

Bitemporal

One fact, two positions in time.

Every fact carries two timestamps — valid time (when it was true) and transaction time (when the firm learned it). The model can rebuild what you knew on any prior day — which is what honest learning requires. You can't fairly grade a decision using information that arrived after it was made.

Canonical

One form. Every variant maps to it.

Firms speak in aliases — the same customer under five spellings, the same concept under three names. The substrate holds one canonical form for every entity and every vocabulary term, and every variant maps to it — a Procore project and an imported bid tab resolve to the same record. Ask in your words; the model answers from one truth.

Polymorphic

Different shapes. One contract.

A score can attach to any subject. A risk can anchor to any trigger. A document can belong to any parent. One mechanism serves every shape — which is why new capabilities land as additions to the model, not migrations of it.

Event Ledger

Everything that happens, in order. Forever.

Every meaningful change writes a typed event to a permanent ledger — who, what, when, and why. Nothing is silently overwritten — 135 event types and counting. The ledger is what makes the model auditable, replayable to any past state, and trustworthy under scrutiny.

The Mirror

Your firm, mirrored fact by fact.

All of it serves one object: a living mirror of your firm — every pursuit, project, commitment, and constraint, connected and current, rewindable to any past moment and projectable forward. Not a report about the business. A model of it.

01c — The Experience: The Illusion of Simplicity

Engineered to Disappear.

An ergonomic, intuitive interface is not cosmetic — it is a core design imperative. The visible layer stays composed and direct: fewer decisions, clearer pathways, stronger defaults, less clutter. The real work is in what you don't see — reconciliation, temporal alignment, identity resolution, exception handling, context assembly across systems. Masked complexity, absorbed below the surface. The interface feels simple precisely because the machinery underneath is doing the difficult work.

Disappearing is not one trick — six kinds of work go away

The menu collapses. What your firm doesn’t use isn’t shown — the interface fits the firm, accordion-like.

Ask instead of navigate. For anything you can say in a sentence, there is no menu at all.

Lessons learned, without the meeting. Knowledge is captured as the work happens — nobody stays late to write the memo.

Handoffs carry their own context. The next person starts where the last one actually stopped — no re-explaining, no archaeology.

The process arrives pre-built. Preset workflows assemble themselves from what you’re already doing — setup work you never do.

“My Plate.” Your work arrives prioritized. You don’t go hunting through the system — it serves you what matters first.

02 — Import Layer

Four Sources.
One Pipeline.

Every external data source — regardless of format, structure, or origin — passes through the same pipeline into the canonical schema. Data doesn't just arrive; it transforms, validates, and becomes operational intelligence.

API

Enterprise Applications

Precon · Projects · Accounting · ERP

Live API connections with delta sync — only what changed moves. Procore, Sage, Viewpoint, and custom endpoints; every firm's schema mapping is persisted.

Semi-structured

Spreadsheets & Exports

Excel · Bid Tabs · CSV · Schedule exports

Upload-based ingestion with intelligent header detection and column mapping — bid tabs normalized across owners and estimators, benchmarked on load.

Unstructured

Documents & Field Notes

RFIs · Field Notes · RFPs · Emails · Submittals

Entity extraction finds the companies, people, and projects inside free-text documents. Originals are stored with provenance; the unstructured history becomes searchable.

External

Market & Intelligence Data

News · Public Postings · Economic Indices · LinkedIn

External feeds resolve to the same canonical companies and people — news, postings, indices, relationship intelligence — every item timestamped for temporal attribution.

03 — Enterprise Integration

Digital Twins.
Real Systems.

This section drills into the first of the four sources above — the enterprise systems themselves. Vertallax is built to work alongside the systems you already run — not to replace them. Preconstruction, project management, and accounting platforms stay where they are; Vertallax creates a synchronized digital twin around them.

From the Shipped Catalog

Application	System Type	Default Integration
Procore	Projects	Procore REST API
Sage Intacct Construction	Accounting	Intacct REST/XML API
NetSuite	Accounting	SuiteTalk (REST)
DESTINI Estimator	Estimating & Precon	File Import
HubSpot	CRM	HubSpot API (v3)
Slack	Communication	Slack Web API
Microsoft Teams	Communication	Microsoft Graph

Seven of the forty systems in the application catalog Vertallax ships — every product typed, every integration door named before day one. All marks are the property of their respective owners.

04b — Model Strategy

Agnostic
by Design.

The models improve constantly — and our infrastructure turns every improvement into upside instead of disruption. Being locked to one model is a risk. Being built to use any of them is protection.

Benchmarked, Not Believed

Our Own Usage Ledger

We run multiple engines and compare cost, performance, coupling, and accuracy on our own usage ledger, all the time — and if one goes down, or a better one shows up, we switch. Where a small model or plain deterministic code beats an LLM, the work moves there. We’ll tell you when the right tool for a job isn’t an LLM at all — which is exactly why you can believe us when it is.

Deployment

Third-Party & Behind the Firewall

Anthropic, OpenAI, or models running behind your firewall — Vertallax is not locked to a single provider. The model serves the firm, not the other way around.

Benchmark

Measured, Model by Model

Every engine runs the same task suites, scored on what’s measurable: accuracy against known‑answer tasks, latency, cost per unit of work, output consistency, and coupling to the job. Benchmarks re‑run as models change — performance is a fact with a date on it, not an impression.

On the Bench Today

Claude

ANTHROPIC

ChatGPT

OPENAI

Perplexity

PERPLEXITY AI

Evaluated continuously, routed by fit — never a dependency. All trademarks are the property of their respective owners; no endorsement implied.

05 — Day-One Value, Cure Time & Time to Plumb

Built for speed.
Cured to plumb.

Day-one value is designed in. The scaffolding arrives preloaded, your documents load from the first day, and answers start flowing the first week. From there, capability compounds on a schedule we measure: cure time — the time to plumb, the point where the Living Enterprise Model is complete and validated enough to support dependable operational intelligence. We call the process truing; we call the end state plumb.

And the pace is deliberate: we grow the data and the rollout in tandem — the model learns the firm while the firm learns the model. No one loads a slab the day of the pour; rushing it is how things crack. Any vendor promising a fully‑loaded model overnight is selling exactly what this industry has learned to be wary of.

Warm the Start

Fifty tables arrive preloaded — configuration, atomic pricing codes, document types, work types — the scaffolding up before your firm enters a single record. Verta Foundations and Kickstart learn the specifics of your operation in plain conversation with your key people. And Smart Document Intake does the heavy lifting: your contracts, your historical RFPs, your documents — loaded from day one. Not a cold start. A running one.

Accelerate the Learning

Half the battle is knowing the dependency map — which items feed which capabilities, in what order, from what origins. We have that map in detail, and purpose-built tools to run it: data-hygiene monitoring and readiness scoring that tell you exactly where the model stands and what unlocks next. A streamlined process, because we know the terrain.

Stage 1 · Kickstart

Construct the initial operating model — it begins before kickoff.

Stage 2 · Cure Time

The key metric: how quickly each functional area reaches operational readiness.

Stage 3 · Plumb

The state where the model is sufficiently complete and validated to support dependable operational intelligence.

Stage 4 · Continuous Optimization

Ongoing refinement as new information, outcomes, and better analytical methods arrive. It never stops.

And philosophically: we don’t block output until it’s perfect. We tell you what we know and how well we know it — every answer carries a confidence level with its dependencies and variables listed. Where the system can confidently manage most of an area — call it 70% — it does. The remainder is surfaced with recommendations on an ergonomic resolution surface.

And 100% of it — including your resolutions — lands in the model, so the confident share grows. The split varies by area; the discipline doesn’t. A high rating means the answer is supported by data. It is not a guarantee — and we’ll be the first to remind you of the difference. We never let AI be more certain than the evidence allows — and we’d rather you weren’t either.

One Model.
One Language.
One Picture.

From Raw Input
to Decisive Action

Six Intrinsic Properties
Built In, Not Bolted On

Engineered to Disappear.

Four Sources.
One Pipeline.

Digital Twins.
Real Systems.

AI Is a Blanket Term.
Ours Is High‑Leverage.

Large Language Models

Natural Language Interface

Agentic AI

Heuristic & Risk Models

Small & Specialized Models

Deterministic Modeling