For the benefit of people
We automate tasks, not professions. Our systems amplify the professional's judgment and leave them the decision — because they carry the responsibility, and responsibility cannot be outsourced to a model.
AI engineering studio · Romania · Europe
The professional decides.
We structure.
We build AI systems for professionals whose decisions carry weight — physicians, lawyers, engineers, teams inside European companies. With cited sources, monitored and audited processing, and data that stays the client's. And with accountability resting, always, on a human being.
01 Mission
AI adopted for the benefit of people. Not a slogan; a technical specification.
Artificial intelligence will enter every profession. The question is not whether, but how: under what rules, with what guarantees, and to whose benefit. Groma exists so that the answer is a good one — for people, for the professions, and for Europe's competitiveness.
Trust through structure
We do not ask for trust; we build it. Authority-ranked sources, mandatory citation, refusal when no source exists, a complete audit trail. Verifiable, not promised.
European competitiveness
Europe will not win the race for the largest models. It can win the race for trustworthy application: AI adopted responsibly, in every firm and every institution. That is where we work.
The name. The groma was the instrument Roman surveyors used to trace straight lines and right angles — direction before speed. Roads set out with it are still in use. The mark: three columns of different shapes holding up the same roof — different professions, one structure. We build on a horizon of generations, not quarters.
02 How we see the world
Civilization runs on professional judgment under pressure.
What separates a working legal system from disorder is not the law itself — it is the trained capacity of practitioners to navigate it. What separates a working medical system from chaos is not the protocols — it is the disciplined mind of the practitioner under uncertainty. The same is true in every field where decisions have consequences.
That is where we work: on orientation — building an accurate model of the situation, fast enough to act. We do not sell knowledge as a possession; we build the capacity to use it under pressure. Our tools do not replace the professional: they tighten the professional's decision loop.
Knowledge well-structured guards life and order. Knowledge poorly applied endangers them.
03 Method
The same discipline, in every product and every implementation.
Five rules we do not negotiate. They are the difference between "we adopted AI" and "we adopted AI and can prove, at any moment, who decided what."
G·1
The human gate
No output with real-world effect without a named human approving it. The system proposes; the professional disposes. Approval is a recorded act, not a reflex click.
G·2
Cite or refuse
Every statement has a source, ranked by authority. When the source is missing, the system refuses to answer — it does not invent. Refusal is a feature, not a defect.
G·3
A complete journal
Every step is recorded: what the model saw, what it proposed, who approved, what changed. Continuously monitored, auditable at any time by whoever holds that right.
G·4
Data stays the client's
Processing in Europe, monitored, with proprietary data. Client data does not train other people's models and does not leave the agreed perimeter. Full stop.
G·5
The craft remains
We automate the task, not the responsibility. We implement AI at the small level of work, with training included, so people keep their competence — and their signature.
04 Work
We do not sell slide decks. The method already runs, in real products.
Three vertical instruments, for three regulated professions, built on the same patent-pending architecture: sources ranked by authority, mandatory element-level citation, refusal when no source exists, expert feedback isolated by tier.
In production · free public good
Custos
Anaesthesia & intensive careA Romanian-language clinical-guideline reference for ICU physicians: answers with verbatim citations from SRATI, ESAIC, AWMF, SFAR, ASA and WHO documents, a hand-curated dosing calculator, case-study workspaces. Verified by physicians, used by physicians — free, permanently.
custossapiens.org In productionLares Pecoris
Veterinary medicineA veterinary medicines reference: withdrawal periods, dosages and official SPCs, with mandatory citation and refusal when the source is missing. It protects the food chain; the veterinarian remains the only one who signs.
larespecoris.vet In developmentLex Vigilans
Tax litigationA fiscal-law reference for lawyers: the Fiscal Code and connected sources, structured by authority tier, cited at element level. Same rule: the instrument orients, the lawyer argues the case.
lexvigilans.roAll three are reference and learning instruments. None of them decides in the professional's place — not by declaration, but by design: no source, no answer; the decision and the signature stay human. We implement the same pattern inside companies, for entire teams.
05 Writing
We publish the argument, not just the product.
Regulation is running behind. Accountability cannot wait.
Frontier models are advancing faster than the rules meant to accompany them. Meanwhile, decisions cannot stay suspended: companies, professions and schools can install human gates of accountability today — and there is already proof that it works.
Read the article →06 Contact
We work directly, with few clients at a time.
If you want to implement AI in your organization — with protected data, audited processing, and people who remain accountable — write to us. We answer, not a form.