QodeX Quantum / AlphaPsi
AlphaPsi acts as a research assistant alongside your researchers. It surfaces connections across the literature, composes candidate designs, recommends optimizations and defends each one with its reasoning, and orchestrates on real QPUs. All in days, not months.
Picks and shovels for the experts you already employ.
AI for quantum AI
§01 What we believe
You can't hire your way out of it. No enterprise quantum team is going to staff its way into the window that matters, so the only move left is to amplify the team you already have.
Quantum advantage won't be a paper. It'll be a better price, a better risk model, a shorter design cycle. The roadmaps put fault tolerance at 2028 to 2029.
Fault tolerance fixes the machine. It doesn't tell you which method wins for your problem. Better hardware will run a validated method faster. It doesn't hand you one.
So the race isn't to the hardware. It's to the method. You get six to eight cycles to find it, in a field where two of three published methods fail on real hardware. Spend them well and you arrive ready. Spend them reading and tinkering and you will arrive late.
Source: McKinsey Quantum Technology Monitor 2026
§02 What it does
01
Published methods, reproduced as working code and benchmarked on real hardware, then composed into designs nobody has published. We keep the result either way: the methods that hold up on a processor, and the two in three that don't for reasons that never made it into a paper. Neither half is in the literature or in any model's training data.
It has done the reading. And the running.
02
Every design ships with the decision logic behind it: the constraints it weighed, the tradeoffs it took, and why this circuit rather than the alternative. Your researchers audit the argument, not just the output, which is what makes them sharper instead of dependent.
The reasoning ships with the design.
03
Every run feeds the corpus. What compounds isn't usage data, it's hardware truth: which methods held up, which broke, on which processor, at what depth, and why. What we learn is which primitives work and which don't, never your problem or your results.
Your science stays yours.
Read the field as it moves, including how hardware actually behaves.
Assemble an algorithm for this problem and this processor, with the resource cost estimated before anything is submitted.
Execute on real hardware. Vendor neutral by design, IBM today.
Feed every run back, the wins and the failures, so the next design starts smarter.
§03 Proof
Training where the field gets zero signal
Our optimizer steers the relative ranking of sampled evaluations instead of chasing a gradient, so training signal holds as circuits reach useful size. In simulation it trains circuits gradient methods fail on every time, at 25 to 30 qubits, on a smaller evaluation budget. On IBM hardware it cold started an exponential barren plateau.
Barren plateaus are quantum's version of deep learning before backpropagation existed.
52 qubits past classical simulation
A PDE forecast on IBM Heron R3 using a circuit no supercomputer can simulate today, plus a 32,000 node unstructured mesh flow on 26 qubits. First hybrid quantum method to handle unstructured meshes, which every real solver uses and every other quantum method avoids.
Paper in preparation, with NVIDIA Quantum and RTX BBN.
Peer review
Nature Communications Physics: the first prediction of chaotic dynamics on NISQ hardware past 100 times the qubits' relaxation time. Plus Physical Review A, Annals of Physics, and preprints on the quantum transformer architecture and application aware error correction.
Six patents filed and growing.
§04 Why work with us
Three led quantum machine learning research for the U.S. Navy and one holds an Oxford physics PhD in quantum algorithm design. They are the scientists behind the patents and the papers. One spent six years inside IBM as global program director helping build their quantum partner ecosystem.
Two exited founders running product and company. Seven people, distributed, and headquartered in Chicago.
Backing
IBM Ventures led the pre-seed. $1.5M raised.
Programs
One of only seven companies to graduate Creative Destruction Lab's quantum cohort from a class of thirty plus. Alchemist Accelerator, Class 41.
§05 Talk to us
Bring a problem your team cares about. We'll show you where AlphaPsi would start, what it would compose, and how it would prove the result on real hardware.
Schedule a meeting with our team