Research
Quantum computing,tested against reality.
Three studies — financial forecasting, genomic sequencing, and molecular classification — each running a quantum method against the classical approach already used in that field, on identical data.
01 — Approach
A result means nothing without a baseline.
Every study here runs the same problem twice — once the way that field already solves it, once with a quantum method — then reports what actually happened, not what we expected to.
03
Completed studies, each benchmarked against classical baselines
82.6%
Average molecular classification accuracy across 5 datasets
~90%
Genomic sequence prediction accuracy, vs ~80% classical
72×
Lower RMSE than the best classical baseline in financial forecasting
02 — Focus Areas
Three domains, three completed studies.
Finance, genomics, and molecular chemistry — each tackled with a different quantum method, and each measured against a classical baseline rather than presented in isolation.
Quantum Finance
Forecasting with QCBM & QMCMC
Modelling market uncertainty instead of just predicting a price.
A Quantum Circuit Born Machine learns the shape of daily AAPL returns — heavy tails included — while a Quantum Dynamical Hamiltonian Monte Carlo sampler (QD-HMC) turns that into calibrated next-day uncertainty forecasts.
QD-HMC's forecasts came in at an RMSE of 0.078, against 5.6 for a naive baseline and 31.3 for ARIMA — on shallow, 4–5 qubit, NISQ-feasible circuits.
Quantum Genomics
Genomic Sequence Prediction
Predictive text, applied to DNA instead of language.
DNA is tokenised into 4-mers (256 tokens, a perfect fit for 8 qubits) and predicted with two quantum approaches: the QD-HMC sampler, and Quixer — a quantum transformer built on LCU and QSVT.
Both reached roughly 90% accuracy, ahead of the ~80% from classical Markov and Transformer baselines, with Quixer's quantum state scaling as O(N) against classical attention's O(N²).
Quantum Chemistry / ML
Kernel Methods for Molecular Classification
A quantum fidelity kernel added to what classical graph features already capture.
Molecular graphs are mapped into a 42-dimensional space — 13 classical topological features plus 29 from a continuous-time quantum walk — then classified with an SVM using a quantum fidelity kernel.
Benchmarked across five datasets (AIDS, MUTAG, PROTEINS, NCI1, PTC_MR) for 82.6% average accuracy, with the quantum kernel improving on 3 of 5 and adding 5.3 points on MUTAG over classical WL alone.
03 — Methodology
How we check our own work.
Public by default
Every active project lives in a public repository from the first commit, not just at release.
Tested outside the lab
Hackathons, competitions, and open benchmarks are how we check whether something holds up outside a controlled environment.
Built on real constraints
We design around actual NISQ noise and actual qubit counts, not idealised simulations.
04 — Open Source
Our code and research live in the open.
Implementations, benchmarks, and write-ups for work like this sit on GitHub alongside everything else we build.
05 — Get Involved
Working on something adjacent?
We collaborate with researchers, labs, and university groups working on quantum finance, genomics, computational chemistry, or applied quantum ML. If that's you, reach out directly.
The hardware is being built by the industry. The open research is on us.