Materials Decoded

The Lab

Everywhere else in AI-for-materials is racing to generate candidates faster. The Lab audits instead: whether the data underneath can carry the weight, and whether an agent's answer is actually good enough to act on. Every artifact here is real, runnable, and honest about what it gets wrong.

Artifact 01

The Trust Benchmark

Can I trust this number? An audit of whether two major materials databases agree, on 344 structure-matched materials, and why they mostly do not.

26%of formation energies agree within the conventional threshold
Explore →
Artifact 02

Battery Electrode Screening Agent

A LangGraph agent that searches real Materials Project electrode data, proposes new candidates when nothing fits, screens them with an ML potential, and grounds every claim in arXiv literature.

2real bugs a live run caught before this was trusted
Explore →
Artifact 03

mfa-engine

A reusable dynamic material flow analysis engine: two real technology systems, Danish wind turbines and the global EV fleet, run through one unmodified engine and connected to real 2023 supply-chain risk data.

105(material, supplier) combinations regression-tested, 0 impossible results
Explore →