Praxis
Processes made by physics, not by trial runs.
Recipes, toolpaths and process windows found by inverse physics, calibrated on each line and validated before it is set up.
Scrap is the teacher today. The world model replaces it.
Expert, CAE, many runs
A process that works after many trial runs. The knowledge retires with the engineer.
Model, plant data, search
A validated process in fewer runs, with error bars. The knowledge compounds.
A process window: a named centreline with its tolerance band, drawn over the deformation field.
Manufacturing first, because a better model pays off in weeks.
Every line it learns makes the next one cheaper.
Calibrated on each line, the model predicts quality inline and corrects deviations live. What it learns is kept: physics shared across lines, each customer's recipes private.
What trial and error costs.
$1.4T is lost every year to unpredicted downtime (Siemens, 2024).
In development. Built against live processes.
Praxis is being built with manufacturers, against their lines and their data.
An automotive assembly station: 100,000+ cycles defect-free.
Footwear robotics: ~90% less setup and reprogramming per design variant.
Four proposals: aerospace finishing, foundry, electronics and process machines.