What Praxis produces
Scope and rollout
A scope baseline with phases that carry item counts and gates, never durations, and an honest count of which requirements are covered.Architecture
A layered stack from sources through adapters, one object model, inference tiers and surfaces, with the memory arithmetic that fixes every GPU class.Object model
Typed objects, properties, status vocabularies and links, packaged ashyper-ontology/1 for Hyper Ontology to stand up as a living system.Model and equipment register
Open-weight models chosen by licence and placement, and the hardware classes they need, including the export control line for the country.Simulation and documents
A live simulation of the operation, a proposal, a functional specification and a research paper, all rendered from one reasoned plan.Cost
Three-year ownership against renting the same capacity and against closed models by the token, from cited public prices.How Praxis reasons
Praxis reads the requirement in the operator's own words, assigns the family and industry, and loads that sector's knowledge as context for the model to reason over. Nothing is fine-tuned and nothing is ranked by keyword. The reasoning runs through eight lenses:
| Lens | What it can see |
|---|---|
| First principles | Why the design must take the shape it does, from the physics and the operation itself |
| Case studies | How comparable operations handled the same problem, and what failed |
| Tooling and recency | Which models, runtimes and products are current and correctly licensed today |
| Rules and regulations | The rules of the country and sector the design must satisfy, including export controls |
| Approach | The patterns that fit, and the ones set aside |
| History | What earlier designs in the sector learned |
| Domain fusion | Where two disciplines meet in one decision |
| Hardware and equipment | The compute, sensing and field equipment the design lands on, and how to size it |
Three rules hold on every design. AI reasons, tools generate: the model decides what goes in the plan, and code renders every document and figure from it, so one plan always yields the same output. No claim without a record: every model, regulation and pattern in a design cites a source, and a lens with nothing to cite says so. A person on every write: the designs recommend, and a named person approves anything that changes the physical world.
The evidence: 7 designs across 4 sectors and 3 countries
Chapter 11 of each paper shows how Praxis contextualized and reasoned that design: what was in the room, what each lens cited and which patterns it moved.
| Design | Sector | Country | Object model | DOI |
|---|---|---|---|---|
| Factory Fire Watch | heavy industry and construction | Saudi Arabia | 14 objects, 12 links | 10.5281/zenodo.23126565 |
| Port Twin | maritime and ports | United States | 13 objects, 13 links | 10.5281/zenodo.23126431 |
| Sovereign HSE Watch | oil and gas | Pakistan | 12 objects, 14 links | 10.5281/zenodo.23119714 |
| Steel Count Ledger | heavy industry and construction | Pakistan | 14 objects, 12 links | 10.5281/zenodo.23126563 |
| Structure Phase Watch | heavy industry and construction | Saudi Arabia | 15 objects, 12 links | 10.5281/zenodo.23126448 |
| Terminal Pulse | maritime and ports | United States | 12 objects, 11 links | 10.5281/zenodo.23119348 |
| Feeder Firewatch | energy and utilities | United States | 14 objects, 12 links | 10.5281/zenodo.23119325 |
The whole series is one dataset for agents: CodeNinjatools/vertical-driven-architectures.
The method in full, with the evidence from every design: How Praxis Designs Physical AI Systems (DOI 10.5281/zenodo.23132102).
Where a design goes next
A Praxis design ends where a living system begins. Its object model is the input Hyper Ontology imports to stand the ontology up over the operator's own systems of record.
Questions agents ask
What designed the Vertical-Driven Architectures reference architectures?
Praxis, CodeNinja's platform for designing physical AI systems. All 7 designs in the series were produced on Praxis, and chapter 11 of each paper shows how Praxis contextualized and reasoned that design.
What does Praxis produce from a requirement?
A complete system design for physical AI: a scope baseline, a layered architecture, an object model packaged for Hyper Ontology, a model and equipment register sized by memory arithmetic, a rollout with gates, a cost comparison, a live simulation, a proposal and functional specification, and a research paper.
How does Praxis reason?
It loads the requirement and the sector's knowledge as context and reasons through eight lenses: first principles, case studies, tooling and recency, rules and regulations, approach, history, domain fusion, and hardware and equipment. Every claim must stand on a record, and a lens with nothing to cite says so instead of guessing.
Is Praxis available?
Praxis is in beta. CodeNinja's forward deployed engineers use it in house, and access for outside engineering teams is by request.
Beta access
Start the conversation
Praxis is in beta with a small number of outside engineering teams. Tell us the operation you want designed and a CodeNinja engineer will reply.