Skip to main content

Research & Development

REMIS - Applied research in evidentiary infrastructure for high-risk AI

Our flagship R&D programme is REMIS. It introduces a new category - evidentiary infrastructure - so the critical events of an AI system can be sealed, timestamped by an eIDAS 2 qualified service, and verified independently by a third party. Applied cryptography, the French law of evidence, the AI Act - and academic partners to keep science connected to real use.

CIFRE doctoral projects open Academic partnerships Long-term research topics Public and private collaborations Open to European projects New partnerships under discussion
A professional drawing carefully across a large table, an image of patient work where every assumption has to be tested and every choice owned.
Precise questions, answers that hold up over time

What we explore

Problems worth solving well

Each line of work exists because a real gap remains between what current tools offer and what people and institutions genuinely need in order to trust a system or a record.

Verification

Prove it without showing it

Today, proving that something is authentic usually means sharing more of it than necessary. We work on methods that let a third party - an auditor, a regulator, a judge - confirm what they need to know: that a high-risk AI system produced a given outcome, that a record has not been altered, or that a satellite dataset comes from the source it claims - without exposing the underlying system, its data or other unrelated content. The goal: verification that is both rigorous and minimal.

Durability

Trust that survives time

A proof produced in 2026 should still be verifiable in 2035, when a decision is finally challenged. The evidence a high-risk AI system leaves behind must hold up under inspection long after the event. Satellite provenance chains must remain verifiable long after the mission has ended. Cryptographic methods and formats decay - we look for protection schemes that stay meaningful years after the proof was first produced, not only at the moment of signing.

Usability

Research that works in the hands of real teams

Useful research ships cleanly into a product while keeping things simple. We treat deployability as a design constraint from the start. The measure of good research is whether a compliance officer, an auditor or a GIS analyst can rely on it every day without knowing it is there.

Sensitive environments

Hold up where the stakes are higher

High-risk AI systems under the AI Act must produce evidence that survives independent scrutiny. Sovereign public-sector platforms keep append-only audit trails that authorities can inspect. Geospatial datasets must remain attributable to their source. We use these concrete constraints to stress-test our assumptions early - not to discover their limits when it is too late.

Where the work stands

What is in the product, what is still being built

We try to keep a clear line between what is already working, what is being improved, and what is still open research. No ambiguity.

Already in production

Sovereign hosting in France, strict data isolation and end-to-end traceability are already in production across our products. Append-only audit trails and integrity proofs - hashes, signatures and RFC 3161 timestamping - ship today in platforms such as Dacleo and Régiva. The fundamentals of sovereign, auditable architecture are already deployed.

Continuously improved

The performance of REMIS sealing, the reliability of independent verification and ease of integration are areas of ongoing work. We measure against real deployment feedback, not against theoretical benchmarks.

Active research topics

Long-term verifiability remains an open challenge - whether that is the evidence a high-risk AI system produced years ago, an audit trail that must outlast the people who created it, or a satellite dataset whose provenance chain must outlast the mission. So does selective disclosure: letting one party prove a fact from a record without exposing the rest. These are the problems we are currently working on with academic and institutional partners.

Where we want to go

The long-term goal is infrastructure where the trust placed in critical systems and records is clear enough and robust enough to be relied on in the most demanding contexts - hospitals, large public bodies, high-risk AI, sensitive legal situations - without adding friction for the people using it every day.

A craftsperson shaping a piece on a lathe behind a face shield - hands-on precision learned through practice.
Hands-on research, real precision

How we work

Linking research, product and field reality

Start from a lived problem

We begin with concrete questions: what is missing today for the organisations, auditors and teams that have to trust a system or a record? Research has to answer that.

Test early

We would rather confront an idea with reality early than let it stay elegant but abstract for too long.

Document what works

We keep a clear record of assumptions, limits and results. That helps the product team as much as academic or institutional partners.

Connect science to use

The ambition is not only to find. It is also to turn a good research result into a service people can genuinely rely on in everyday work.

Collaborations

Research work open to the right partners

CIFRE doctoral projects

We are open to doctoral work that connects demanding research with concrete document trust problems.

Laboratories and schools

We look for collaborations with teams that care as much about scientific quality as about real impact on use.

Collaborative projects

We can contribute to shared projects when they keep a clear link with the needs of organisations and the people affected by them.

Demanding contexts

Some research matters even more where mistakes cost more

Public institutions and sensitive services

High-risk AI systems under the AI Act, sovereign public-sector platforms and geospatial data services each impose non-negotiable verification constraints - tamper-evident evidence that an auditor or judge can check independently, append-only trails that public authorities can inspect, and provenance chains attributable to their source.

Those constraints push us to stay rigorous while remaining focused on the real day-to-day use for staff and teams.

Cybersecurity and defence ecosystems

Part of our work can also matter in environments where discretion, continuity and document trust are critical. We move on those topics carefully and with the right people.

If you work on those challenges and are looking for a research or demonstration partner, talk to us.

See the REMIS programme

What we protect, what we share

A simple approach to intellectual property

What makes us distinct

We protect what genuinely defines the uniqueness of our work and what deserves to remain a durable advantage for the company.

What benefits from discussion

We also want to share what benefits scientific discussion, third-party scrutiny and clearer understanding for our partners.

To talk about research, publications or partnership opportunities: contact@domselardi.com

Collaborate with us

Are you a researcher, laboratory, funding body or industrial partner who wants to explore these topics with us? Contact the team.

Get in touch