One engine.
Infinite experiments.
An LLM orchestrator paired with an LQM simulation layer: a closed, self-improving loop evaluating millions of candidates while you sleep.
It writes the science.
Generates novel, falsifiable scientific hypotheses grounded in domain knowledge, and translates each into a formal Experiment Specification for the quantitative engine to evaluate.
It runs the physics.
A physics-aware ML model predicting domain properties (formation energy, bandgap and more) in milliseconds per structure. Every prediction carries a calibrated uncertainty score.
The 5-step discovery engine
Each colour is a stage. Each loop compresses months of bench time into minutes.
Define
You set the domain, the target property envelope, and the performance threshold. The engine takes it from there.
Hypothesise
The LLM generates falsifiable scientific hypotheses and compiles each into a formal Experiment Specification.
Simulate
The LQM evaluates thousands of candidates in parallel, at millisecond speed per structure, with uncertainty on every prediction.
Validate
Anomalies escalate to an exact DFT oracle. The LQM retrains on the labelled results, and every loop makes it sharper.
Deliver
A ranked, confidence-annotated discovery report lands, ready for experimental handoff.
See it deployed live.
NexCon-03 is already running this loop against compound semiconductors.
Explore NexCon-03 →