Story
PINNeAPPle exists because the working had to be visible.
AI models for engineering are easy to sell and hard to believe. PINNeAPPle started as a personal library so its author would not have to take anything on trust: every result carries the reference it was checked against.
Why
Why verification first, and not just a model
A PINN can hit its target while violating the equation along the way. The training loss looks healthy and the physics residual sits hundreds of times over the threshold — which is exactly what we found in an already-trained model from a real client.
If it has no number measured against a reference external to training, it does not go on screen. That holds for a README, for a slide and for this landing page.
Saying "no" is part of the product. In one rover-mobility case the classical numerical method was faster and more accurate than the network — the recommendation was the classical method.
How
How it is built
31 modules on PyPI, Apache-2.0, 248 examples with checked output. The base is public because a verification tool nobody can audit verifies nothing.
SSQN Helmholtz, LBM vortex shedding, LPM design space, latency and I/O. No internal benchmark presented as proof.
Each with a README, numbers and stated limits. They are a starting point for your own product, or proof that the base holds up on the real case.
Team
Who builds it
Physics, industrial engineering and software on the same team. The rule is the same as before: if it has no measured number, it does not go on screen.
A physicist by training, he combines physics, AI and engineering to take scientific models into real applications. He created PINNeAPPle and works on AI applied to real engineering and industrial problems in Brazil. His background includes climate, agribusiness and solar astrophysics.
LinkedIn →Bring your case
Got a simulation that is slow, expensive or hard to verify?
Send it over. We come back with a closed scope and a measurable success criterion — or we tell you the classical method does it better.