Programs, not sliders
Leif can change topology, branching, channel routing, lattices, and manifolds by writing executable geometry programs.
Vinland Labs / Leif
An AI model for autonomous hardware invention.
We build Leif. R&D teams give it a hardware problem, operating conditions, and manufacturing constraints. Leif writes geometry programs, evaluates new designs, and learns what works.
Loading generated geometry
Vortex-core heat exchanger
Voxel-native / 3 mesh bodies / drag to inspect
Current research
01
Cooling & flow hardware
02
Voxel / SDF geometry
03
Program synthesis
The problem
The design space is too large for humans to search.
Hardware teams can imagine more topologies, channel layouts, materials, and manufacturing processes than they can evaluate. Conventional optimization narrows the problem to a human-defined shape family before the search begins.
Conventional search
Human-defined parameter box
Unsearched structures
New geometry programs
Explored
Leif search
A vast design space is only useful if the system can learn structure inside it.
Physics closes the loop
Candidate designs are scored with simulation, experiments, or learned surrogate models against real engineering objectives.
Each run leaves knowledge
Successful patterns are compressed into reusable building blocks for future design problems.
How Leif learns
Few examples. A growing design language.
Leif combines program synthesis, library learning, and deep learning. It searches for programs that generate useful hardware, then turns repeated solutions into concepts it can reuse.
01
Program synthesis
Generate causal designs
Leif writes compact programs that control how a geometry is constructed, not just where individual mesh points move.
02
Library learning
Compress what works
Repeated geometry patterns become named, reusable abstractions: a learned library of hardware design rules.
03
Deep learning
Choose where to search
Learned models guide program proposals and predict which candidates deserve expensive simulation or physical testing.
The result
A model that improves across problems.
Traditional generative design optimizes one shape for one brief. Leif carries useful concepts forward, then reuses and recombines them on unseen hardware problems.
Starting with cooling
Real inputs. Manufacturable outputs.
We work directly with R&D teams on heat exchangers, cold plates, manifolds, and internal flow channels.
You provide a baseline design, operating conditions, target metrics, and manufacturing constraints. Leif returns ranked voxel/SDF candidate geometries, predicted performance, and a design report for downstream CAD, simulation, or manufacturing.
Maximum temperature
Heat transfer
Pressure drop
Mass
Packaging
Manufacturability
Design partners
Bring Leif a hard hardware problem.
We are working with R&D teams on tightly scoped design studies. Tell us what you are trying to improve and what the design must obey.
Work with Vinland Labs