Pakistani technology company CodeNinja has released a set of open reference architectures for physical AI through its CodeNinja Atoms platform, giving engineers free access to full system designs for industries such as energy, oil and gas, ports, rail, manufacturing, heavy industry, aviation and agriculture.
The blueprints, published on November 13 at codeatoms.ai, are designed for real operations in the United States, Saudi Arabia and Pakistan. The company says the aim is to let organisations deploy physical AI that stays under their own control.
What the blueprints include
Each architecture sets out the full design of a system rather than just a model. According to the company, every blueprint covers:
- the design logic behind the deployment
- what the system needs to sense
- where each AI model runs
- an object model that connects the operation’s existing systems
- the open-weight models and hardware required
- three-year costs
- the named person who approves each action
The library is released under the Creative Commons Attribution 4.0 International licence, which means engineers and companies can study, reuse and adapt it for free.
Why CodeNinja is publishing it
CodeNinja argues that many organisations use AI without owning the design logic, model choices or operating knowledge behind it. When those stay with an outside vendor, the organisation cannot properly inspect, change or extend the system it depends on.
Muhammad Umar Bilal, co-founder of CodeNinja, said an organisation cannot govern a system it cannot see. By publishing the design, object model and data behind each architecture, he said, engineers and operators can inspect the systems, adapt them and build them on infrastructure they control.
“We want every engineer to build on these designs,” Bilal said, adding that when an operator is ready for its own system, CodeNinja’s forward deployed engineers will build it inside the operator’s own premises.
Praxis design engine
Alongside the library, CodeNinja is offering developer access to Praxis, a blueprint engine for physical AI based on published research and industry standards. The company says the hardest part of physical AI is usually the system design, not the model, and that design expertise is scarce.
A user describes one operational requirement in plain words. Praxis then returns a complete system design covering sensing, model placement, the object model, hardware, three-year cost and the approvals each action needs, with the reasoning behind every choice. A live simulation shows the operation running on the design before anything is built.
Praxis uses a recursive learning loop. Each design, benchmark run and engineer correction is saved to memory and used for the next requirement.
Benchmark results
CodeNinja scores Praxis on its own public benchmark, the Physical AI Design Index (PADI). The benchmark gives Praxis and a frontier AI model the same short requirement and checks both designs against the same criteria across 13 industries.
In the latest blind evaluation, called Loop 22 and run on October 16 with seven unseen tasks, Praxis passed 48.5 percent of checks, against 28.4 percent for frontier AI working alone. Praxis led on all seven tasks. The company says earlier loops used different tasks, so they show a trend rather than a direct comparison, and that scores are still far from the maximum.
Where to find it
The full architectures are available at codeatoms.ai. The complete library is also published as an open dataset, CodeNinjatools/vertical-driven-architectures, on Hugging Face, with tables for designs, object models, model choices and cost lines. Each paper has its own DOI on Zenodo. Engineers can apply for Praxis developer access through the same website.





