Experiment
terrain-builder
activeNatural-language to print-ready wargaming terrain. The user describes a building ("L-shaped townhall, three floors, gothic, balcony on the second floor"); the system selects pieces from modular STL kits, generates a layout, gets human approval, and produces split-for-printer STLs with alignment features preserved.
How much of a 3D-modelling pipeline can you hand off when the inputs are kit STLs full of clip slots and cone pins? The messy mechanical bits (boolean operations that don’t shred the connectors, splitting at architectural seams, orienting for a supportless print) are exactly the kind of work I’d rather push to an agent than do by hand.
Live questions:
- Approval gates: where in the pipeline does a human “looks right” check actually earn its keep, and where is it just noise
- Connector preservation through CSG: keeping clips and pins intact through merges and splits, treated as a hard correctness rule
- Cross-kit compatibility: whether onboarding a new STL kit can become a one-time automated job
Field notes
No posts tag this experiment yet. When a post does, add experiments: [terrain-builder] to its frontmatter.