NEUTRINO IAG
Patents filed · 2026

Solve the real world
off the grid. Universal evaluation kernel for the treatment of instability resolution and optimised non-linearity.

NEUTRINO replaces brute computing power with entropic arbitration. An instability score decides, in under 15 ms, whether a physical state belongs to low-power classical processing or to the quantum kernel. Locally, under 5 W, behind an air-gap barrier.

air-gapped enclosurelatent space of the kernel
instability score Ω0.00
classical regime · under 5 W
0documented applications across 14 sectors
0patent families, 1 core and 51 applications
0decision latency of the entropic sieve
0power draw of the decision unit on ASIC or FPGA
0cumulative addressable market across sectors
0total cost of ownership against cloud offerings
The industrial lock

Three imperatives no architecture held together.

Fluid turbulence, plasma instability, protein folding, chaotic dynamics: solving these systems currently means sacrificing at least one of three strategic pillars. NEUTRINO is designed to sacrifice none of them.

Sovereignty

Leading hybrid architectures rely on cloud connectivity. Industrial and defence data leaves the enclosure, and a true air gap becomes structurally impossible.

Frugality

The local alternative demands power-hungry GPU clusters or non-transportable quantum computers, out of reach for an operational site.

Independence

Dependence on monolithic chip foundries is a geostrategic vulnerability that neither software nor contracts can correct.

Architecture

Four pillars, a single arbiter.

Pillar 01

Geometric compression

Physical laws are mapped into a latent space that sharply reduces the number of state variables while preserving the relations between them.

Pillar 02

Entropic evaluator

An asynchronous sieve computes a global instability score in under 15 ms and isolates the singularities that deserve special treatment.

Pillar 03

Evolutionary optimiser

A mutation engine explores a very large configuration space and converges the architecture toward the shape best suited to the problem at hand.

Pillar 04

Structure-informed distillation

Correlations discovered by the quantum kernel are transferred to a local classical model under strict conservation-law constraints.

Ω  ⟶  τ An instability score Ω is compared against a critical threshold τ. Above it, the fast controller engages the quantum kernel within nanoseconds. Below it, processing stays classical and low-power. The QPU is not a permanent engine, it is a singularity accelerator. The exact composition of the score and its calibration are trade secrets.
A B C D LOOP 4 phases
  • A
    CompressionThe physical state vector is projected into the compressed latent space of the kernel.
  • B
    Ω evaluationThe instability score is computed and arbitrates between the classical and quantum paths against threshold τ.
  • C
    Self-optimisationArchitecture search inside the latent space, guided by the mutations of the optimiser.
  • D
    DistillationTransfer to the local classical model, under the physical constraints of the system being solved.
Intellectual property

One foundational base, fifty-one derivatives.

The patents filed in 2026 protect the hybrid artificial general intelligence architecture and its resolution method: the system, the method, the uses and the program product.

Around that base, fifty-one application patents cover the kernel building blocks and their sector variants. Each block remains separately licensable, which turns the portfolio into a platform.

Portfolio

Hover a cell to place the matching patent. The amber cell is the base, the following ones are the technology building blocks and then the sector variants. Detailed titles are shared under a non-disclosure agreement.

Mapping

A universal kernel, not a sector tool.

Every application reduces to the same problem: a non-linear system whose instability must be predicted, inverted or optimised without connectivity. Forty applications are prioritised and break down into ninety sector variants.

Feasibility

What runs today, what waits on hardware.

Application setHorizonHardware requiredMain obstacle
Digital twins, power grids, sovereign intelligence watchShortClassical on FPGAData access and field integration
Quantitative finance, minerals, weather, geophysicsShort to mediumClassical and light QPUQuality of historical crisis data
Autonomous maritime, biotech, seismic, roboticsMediumNISQ QPUDecoherence and noise mitigation
Fusion plasma control, hypersonic guidanceMedium to longAdvanced NISQ QPUNanosecond latency in a real-time loop
Full 3D turbulence, ambient superconductivityBreakthroughFault-tolerant QPUScaling the architecture

Breakthrough applications require better hardware, but also that the architecture holds its promises beyond synthetic cases. That scaling demonstration is part of the programme, it is not presented as settled.

Sovereignty and ethics

The air gap is not a deployment option.

It is an architectural property. Sensitive data never leaves the enclosure and the archive is encrypted with lattice-based post-quantum cryptography. The design answers the confidentiality requirements of defence, energy and biotechnology.

Several capabilities on the breakthrough tier are inherently dual-use. The portfolio is therefore split between a defensive sovereignty base, freely commercialised, and a restricted compartment of sensitive capabilities subject to authorisation.

The applicable export regime is documented before any demonstration to a foreign buyer, and an internal ethics committee oversees applications touching opinion, living systems and human cognition.

Join the team

We are hiring rare profiles.

The team is deliberately small and works in Paris on local, disconnected systems. We look for people who can hold the rigour of a laboratory and the pace of an industrial programme: PhDs, senior engineers, and final-year internships on defined subjects.

Details of the work are shared once a non-disclosure agreement is signed. Some positions require a nationality or a clearance compatible with sovereignty projects.

Applied mathematics

Differential geometry, dynamical systems, non-linear partial differential equations, information theory. PhD preferred.

Quantum computing

Variational circuits, error and noise mitigation, NISQ and neutral-atom platforms, porting algorithms onto real hardware.

Machine learning

Structure-informed models, distillation, optimisation under physical constraints, model reduction for embedded targets.

Embedded systems

FPGA and ASIC, real time, very low power, design of isolated and hardened processing chains.

Security and cryptography

Post-quantum cryptography, security of disconnected systems, randomness generation, hardening of sovereign archives.

Scientific software engineering

Python and C++, high performance computing, numerical validation, industrialising a compute kernel and its test suite.

Signal processing

Sensor physics and inversion: geophysics, radar, underwater acoustics, imaging and reconstruction.

Business development

Selling complex programmes to public and industrial buyers, project financing, intellectual property management.

Send an application

Name the field you are applying for and attach a CV. Speculative applications in adjacent skills are read.

Investors

The pitch deck is available on request.

NEUTRINO IAG is opening its funding round to deeptech investors, sovereign funds and industrial partners. The deck covers the patent portfolio, the execution plan, the first committed use cases and the financial projections.

None of this material is published. It is shared by name, after an initial conversation and a signed non-disclosure agreement.

Request the pitch deck