Some context. I'm a web designer from Vilnius. No math degree. Over the past few months I built a geometric synthesis engine (~7,700 lines of Python) that encodes concepts as complex phasor vectors on an E8 lattice and discovers structural relationships through algebraic binding. The LLM doesn't reason — it's just a vocoder that names what the geometry produces. All structure is deterministic. Same input, same output, every time.
The mode I want to show is Reverse Genesis. You give it a goal, it decomposes the structural prerequisites — what has to be true before this outcome is possible.
I pointed it at e^(iπ) + 1 = 0. It produced 27 prerequisites.
Things like "Spinor Sign Flip — a half-rotation in a hidden spin space to align phase factors for cancellation." That's... mathematically correct. e^(iπ) is a π rotation, which in spinor algebra is literally a sign flip.
"Root Collapse Path — collapsing the exponential's infinite branching into a single real-valued negation." That's the Taylor series converging to -1. None of the 27 were wrong.
But here's what actually surprised me. I ran the same engine on alpha-synuclein aggregation in Parkinson's — the protein that's defeated drug design for 30 years because it has no stable shape to target. 386 nodes came back. Three of the decomposition paths don't match anything in the current pharma literature. And it independently found liquid-liquid phase separation as a critical layer, which the field only confirmed in 2020. The engine had zero biology context. It got there from geometry alone.
I should be honest about what it doesn't do. It doesn't prove theorems. It doesn't solve equations. It identifies structural prerequisites — what has to be geometrically true before something is possible. Sometimes that's profound. Sometimes it's the obvious answer dressed in geometric vocabulary.
I'd estimate 30-40% of findings are genuinely non-obvious, 40-50% are correct but known, and 10-20% are noise. I'm still calibrating.
I'm looking for hard targets. The weirder the better. What would you want to see it decompose?
Some early papers at omuo.io/research. Patent pending.