A focused research company with one problem

Mindverse Computing is building the agentic quantum digital twin learned from BCI data. We narrowed to that deliberately, from a broader remit.

Where we are today

We are at the research-to-product stage: the Quantum Virtual Mind (QVM) — the quantum–classical core — is built and benchmarked, manuscripts are in preparation, and the applications are moving into development.

1
Core model built

Benchmarked in house against a matched classical baseline on public EEG data (PhysioNet EEGMMIDB), with a mixed verdict reported in full.

3
Manuscripts

In preparation: the foundation model, the healthy-variation atlas, and the longitudinal clinical record.

3
Applications

Superlearning, restoration, and clinical research — three applications drawing on one core rather than three separate builds.

4
Market routes

Research tools, education, wellbeing, clinical research — sequenced by difficulty, all built on the same state reading.

Four things that distinguish this

A defensible structural claim

The QVM is constructed so it cannot produce physically impossible states. A structural property rather than a benchmark — the part of our approach least dependent on results still to come.

Evidence we report in full

Three manuscripts in preparation, and an early benchmark we report in full — the measures we lead on and the ones conventional methods still lead on. Both are in preparation.

Genuine compounding by design

Both proposed directions would use the same recordings, preprocessing, model, and agents. An architectural advantage, with the shared machinery already built into the design.

A sequenced route to market

Research tools first, then education, then clinical research — ordered by how long each takes to validate, not by how large each looks. Research groups already hold the recordings and can evaluate a tool in months; clinical work needs years of longitudinal data before anything can be claimed. Entering in that order means revenue and evidence arrive before the expensive validation begins.

BUSINESS MODEL

How this becomes a business

We do not intend to sell headsets or run a consumer app. The plan is to be the layer between a customer’s recordings and whatever they want to do with them: their data lands on our platform, a state model is fitted to each of their subjects, and they reach it through an API. One core, billed three ways.

Layer 1 · Ingest

Data platform

Customer recordings land in a managed pipeline — consented, provenance-tracked, and conditioned into a form a model can be fitted to. Subscription per organisation, scaling with volume of data under management.

Charged whether or not a model is trained. The unglamorous part, and the part that creates switching cost.

Layer 2 · Train

Model fitting

A Quantum Virtual Mind fitted per subject and refitted as new sessions arrive. Priced per model trained, or bundled into an enterprise tier for customers running continuous cohorts.

The compute-heavy step. Margin improves as the fitting pipeline matures.

Layer 3 · Serve

Inference API

The state model exposed as an API, metered per call. This is where the three applications diverge — same core, different endpoints and different buyers.

The recurring line, and the one that grows with a customer’s own usage rather than with our headcount.

Three endpoints, three buyers

Superlearning API

Learning platforms and training programmes call it for readiness and pacing. They keep their content and their user relationship; we are the layer behind it.

Restoration API

Meditation, sleep, and stress apps call it to select and pace a session from measured state instead of a preset track. Highest call volume, lowest price per call.

Clinical research API

Research groups call it for longitudinal trajectories against a subject’s own baseline. Lowest volume, longest contracts, and the strongest reason to stay.

None of this is generating revenue yet — the model is built, the agent layer is in development, and the applications are proposals. This is the structure we are building toward, not a description of current operations. Pricing and projections are in the investor materials.

Where things stand

The state modelBuilt, early testing
Independent validationNot yet
ManuscriptsThree, in preparation
Agent layerSpecified, being built
SuperlearningProposal
Disorder studiesProposal
Clinical deploymentNone — by design

What exists, and what does not

The state model is built and has had one early internal test on public data. The agent layer is specified and being built. All three applications are at proposal stage, specified in draft manuscripts and moving into build.

What you would be backing

We would rather tell you which parts exist than imply the whole stack is finished. If you are evaluating this, you are backing a technical approach and a team at the point where the science is proving out.

See where the work stands →

Why now

Recording is solved; interpretation is not

Consumer and research BCI hardware has improved steadily, and the number of recordings being collected grows every year. What has not kept pace is the ability to say what those recordings mean for a specific person. That gap is where we sit, and it widens as acquisition gets cheaper.

Agentic systems became practical

Reliable tool-using AI agents are recent enough that wrapping a scientific model in an agent layer was not a realistic architecture three years ago. The model and the agents are each ordinary now; the combination applied to brain dynamics is not yet crowded.

Interpretability became a requirement

Regulators increasingly require that a model’s behaviour can be explained and audited. A model built from named components starts from that position.

Materials

Detailed materials — technical roadmap, market sizing, financials, and the current manuscript drafts — are shared with qualified investors on request, under NDA where appropriate. We are happy to walk through the research in depth, including the results that did not go our way.