I built Grounded Discovery Labs, a system that takes a science or engineering problem statement and produces a ranked portfolio of candidate concepts, along with decisive experiments and kill criteria.
This is not just a thin AI wrapper around a model. The point is not simply to generate ideas, but to force candidate concepts through a structured workflow that pressure-tests them, repairs weak points where possible, and tries to separate stronger directions from noise and hallucination.