Company

We want new materials in days, not decades

Batteries, catalysts and chips are limited by the materials we know how to make. Forgite exists to shorten the time between a target and a sample that meets it.

Abstract

Stage
Pre-launch
Focus
Batteries, catalysts
Domain
forgite.com
Team
Hiring

01 · Why we started

Accurate enough was always too slow

Computational chemists have had accurate methods for decades. What they have not had is the speed to apply them to more than a few hundred candidates, or a clean way to connect predictions to what happens in the lab.

ML interatomic potentials changed the speed. What is missing is the industrial part: models that report their uncertainty, a loop that decides what to make next, and a record of every step. That is what we are building.

02 · Principles

How we work

01

Measure first

We publish errors before claims.

02

Lab time is sacred

Every sample we propose should be worth making.

03

Your data stays yours

Opt-in sharing only, always.

04

Write it down

If it cannot be rerun, it did not happen.

03 · Roadmap

What we plan to build, in order

Phase 1

Battery screen

Equiva battery pack, Dynara stability and conductivity, Inventa substitution search.

Planned

Phase 2

Closed loop

Synthex with partner cloud labs; Mentra with literature grounding.

Planned

Phase 3

Catalysts

Catalyst model pack, adsorption calculators, on-premises release.

Planned

Phase 4

Generative design

Inverse design from targets; process digital twin for scale-up.

Planned

04 · Careers

Open roles

ML research scientist, interatomic potentials

Equivariant GNNs, uncertaintyRemote or hybrid

Computational materials scientist

DFT, MD, batteries or catalystsRemote or hybrid

GPU software engineer

CUDA, Warp, TritonRemote or hybrid

Platform engineer

Kubernetes, data provenanceRemote or hybrid

Lab integration engineer

Instrument data, cloud-lab APIsHybrid

Send a short note and something you have built to careers@forgite.com.

05 · Stack

What we build with

ModelsPyTorch, cuEquivarianceEquivariant potentials
SimulationWarp, ASEMD and properties
ServingTriton, NIMGPU inference
DataRAPIDS, ParquetCampaign results

Next step

Work on materials that do not exist yet

We are a small team. If the problem on this page is the one you want to work on, write to us.