We build specialised AI agents for operational decisions.
Pragmatica specialises in agentic AI built on privacy by design and on the principles of the EU AI Act. Our agents are super-specialised, deterministic and non-generative: each one does a single operational job, follows explicit rules and needs few computing resources. qb-Agentic is the core that coordinates them; qb-Plant is the first product built on it.
From signal to approved, recorded decision.
An absence, a message or a plan change reaches the agents.
The context is modelled as a graph: people, skills, certifications, machines, shifts, rules.
Rules are checked deterministically; agreements are stored as rules, cited by article.
The situation is re-solved at event time and the feasible options are ranked.
A person approves, edits or rejects.
Decision, rule, author and time are saved.
Small agents. Clear rules. People in control.
Super-specialised
One agent, one job: detecting a disruption, checking a rule, ranking replacements.
Deterministic
Same input, same output. Every rule is applied exactly and cited by article.
Non-generative
Agents select and rank among feasible options; they do not generate text, answers or decisions of their own.
Light
Few computing resources, so agents run in the EU cloud or on the client’s own servers.
Privacy by design
Client data is anonymised, kept separate from the decision engine and under the client’s control.
Human in the loop
No agent executes a sensitive action without a person’s approval, and every step is logged.
For decisions about people, plausible is not enough.
Decisions on working time, safety and contracts must be correct, repeatable and explainable. Generative models can produce answers that sound right but are not, and they need large data centres. Our agents work inside explicit rules, give the same result every time and can run on modest infrastructure. This is how we align with the EU AI Act’s requirements for human oversight and record-keeping in high-risk systems.
- Produces plausible text
- Output can change for the same input
- Large models, large data centres
- Applies explicit rules, cited by article
- Same input, same result
- Small agents, EU cloud or on-premise
Standard building blocks. Pragmatica decision layer.
Selected optimisation and disruption-management methods.
Workforce-software foundation already built.
Our AI platform project is co-funded by the European Union.
“Artificial Intelligence platform for the digital transformation of SMEs in Friuli Venezia Giulia” — PR FESR 2021-2027 Friuli Venezia Giulia, call a3.4.3 “Interventions supporting entrepreneurship”. CUP D92D25000230007.
Project details