Building the infrastructurefor quantum development.

Syndrome-Net treats decoding and calibration as something an agent can adapt to, rather than a fixed set of rules tuned for one noise profile.

We train the decoder with reinforcement learning — TITANS, Proximal Policy Optimization (PPO), and Soft Actor-Critic (SAC) — instead of hand-tuned heuristics. It supports surface codes, qLDPC, and colour codes, and is built to hold up as NISQ noise drifts rather than staying fixed.

QuantumForge is the engine behind our more compute-intensive work — built to close the gap between high-level circuit code and the execution speed needed to iterate on it.

Written in Rust, it simulates 26-qubit statevectors on consumer hardware at roughly 4,000× the speed of a typical Python implementation. That's the difference between waiting on a cluster and iterating locally on algorithms like GQE and VQE.

This work pairs classical machine learning with quantum workflows to tackle high-dimensional, non-convex optimisation problems that neither approach handles well alone.

We study how a given device's noise profile limits which algorithms are actually viable, and apply that to combinatorial optimisation and market simulation. Hackathons and open-source contributions are how we test this against real problems instead of benchmarks we picked ourselves.