Artificial intelligence and data
- Home
- Artificial intelligence and data
We don’t talk about artificial intelligence as a promise. We use it on our own systems every day, and this is what runs in production.
What we have built
Since 2024 we have brought artificial intelligence inside our management platform and into the platforms of the research projects we take part in. It is not a pilot and it is not a demo: these are systems that work every day on our data and on the data of the European projects where we are the technology partner.
| What | What it does | How much it works |
|---|---|---|
| Documentation assistant (RAG) | Answers questions about manuals and procedures, citing the document and paragraph the answer comes from | 210 pages indexed, 1,025 retrievable vector chunks |
| Client dossier enrichment | Reads the information available about a company and prepares a structured brief for whoever has to write a quotation | 6,873 dossiers enriched |
| Automated document analysis | Extracts structured data from emails, orders and attachments, and proposes a classification | one of eight active contexts |
| Image recognition | A computer vision model trained to recognise litter in beach imagery, born out of the REMEDIES Horizon project | research, see below |
| Predictive models on sensor data | Estimates water chemistry parameters that are not directly measured, starting from eight probe readings | 11 estimated out of 19 |
All of it sits on an abstraction layer that does not tie us to a single vendor: the system routes each request to the most suitable model — we use Anthropic’s Claude, Google’s Gemini and OpenAI’s models — and the configuration changes without touching the code. It is a choice about independence before it is a choice about performance.
The principle, which is a board resolution and not a slogan
Artificial intelligence proposes, explains, prioritises and predicts. It does not decide on pay, discipline, promotion, dismissal or access.
One consequence concerns anyone using our workforce management software directly: we do not do individual scoring of employees, and we will not. Time and attendance systems are there to count hours, not to judge people.
A second consequence concerns everyone: every AI output is a suggestion, not a final decision. Wherever our software proposes something — a classification, a priority, an answer — a person confirms it, and the source of the suggestion is visible.
Read the full page on transparency and the AI Act →
What we can do for you
We learned by building for ourselves, and now we do it for clients too. We don’t sell models: we sell the work of understanding where AI actually helps and getting it into production inside systems that already exist.
Six ways to engage, from a half-day assessment to a part-time engineer on your project.
Go to consulting and AI engineering →
Where AI meets sensors
The most advanced part of our research is not about text: it is about data coming from the field.
acquamArIna measures water quality continuously with eight probes and estimates eleven further parameters with a model trained on laboratory samples. The result is a single index, a public dashboard and an open API. Discover acquamArIna →
Karst Firewall combines a sensor we designed ourselves with predictive algorithms, to monitor wildfire risk and groundwater quality across the cross-border Karst region. Go to research and innovation →
Beach litter recognition was born inside the REMEDIES Horizon project: a computer vision model trained on images collected through a citizen science app, and today the basis of a platform project for environmental research. ⚠️ This is ongoing research: it is not a product on sale.
Questions we get asked
Do you use AI on our employees’ data?
No. Our time and attendance and access control modules do not apply artificial intelligence to individual worker data, and the company resolution forbidding it is public. Where AI is present in our products, it works on technical documentation, commercial text and sensor data.
Is our data used to train models?
No. We use vendor models through programming interfaces that do not allow content to be used for training, and client data stays on our systems. The list of vendors is published on our transparency page.
Where is the data held?
In our own datacentre, in a former bank vault in Trieste, with Italian national cybersecurity agency (ACN) cloud qualification. Our cloud service data is hosted there; for AI features, the vendors involved and the data flows are listed on our transparency page. Go to the datacentre page →
Does the AI Act apply to you?
Yes, and we have dealt with it. Some access control and workforce management modules fall under Annex III of Regulation (EU) 2024/1689, which applies from 2 December 2027. We have classified our modules by risk level and completed the AI literacy training required by Article 4 for everyone in the company.
Do you consult for organisations that are not already your clients?
Yes. AI consulting is a standalone service and does not require you to run our systems.
In short
Eight AI contexts in production · 1,025 retrievable document chunks · 6,873 enriched dossiers · a computer vision model born from a Horizon project · eleven water quality parameters estimated by model · no employee scoring, by board resolution.