DOLMIR

Reliability

Intelligence is only useful when you can control it.

Every AI vendor will show you what happens when it works. This page shows the other case — the one that decides whether a tool stays in use or gets abandoned — and the principles the system is built on to stay under control.

  • VERIFIED
  • CONFLICT DETECTED
  • DECISION REQUIRED
  • UNDETERMINED
  • HUMAN APPROVAL

How it stays under control

5 in the product · 3 principles

  1. Evidence

    In the product today

  2. Verification

    In the product today

  3. Human approval

    In the product today

  4. Controlled actions

    In the product today

  5. Declared limits

    In the product today

  6. Traceability

    Design principle

  7. Permissions

    Design principle

  8. Security

    Design principle

What happens when the system does not know

  1. I DON’T KNOW

    The system recognises the limit

    A code that matches nothing in the catalogue, a missing quantity, an unreadable document.

    01

  2. STOP

    The process halts

    No plausible value is produced. Nothing is produced.

    02

  3. PERSON

    The case is assigned

    With what is missing and why, not with a generic error message.

    03

  4. DECIDES

    The decision goes back to a person

    Whoever knows the customer completes, corrects, approves. The system records the case; it does not forget it.

    04

DOLMIR does not manufacture technical certainties it does not have.

A language model has no native way of saying “I don’t know”: the most fluent answer is always available. That behaviour has to be built against the model, with thresholds, evidence and explicit gates. It is the part of the work nobody sees in a demo, and it decides everything else.

How it stays under control

Eight principles. Five already in the product, three by design.

  1. In the product today

    Evidence

    Every extracted value carries its source: the line of the email it comes from. A value without a source does not exist.

  2. In the product today

    Verification

    Every value is checked against what the company knows — customer records, item catalogue — and the result is stated: matches, does not match, cannot be determined.

  3. In the product today

    Human approval

    No reply goes out without a person’s yes. The system prepares and waits.

  4. In the product today

    Controlled actions

    The system performs only the actions it was built for: reading, verifying, preparing. It does not write to the management system, does not send on its own, does not calculate prices.

  5. In the product today

    Declared limits

    What it does and does not do is written down before signing, and repeated on this site. A system that “does everything” is a system to distrust.

  6. Design principle

    Traceability

    What was read, verified and decided, and by whom, must stay written down and available to consult. That is how we design every system we build.

  7. Design principle

    Permissions

    Every system accesses only what it needs, read-only where reading is enough. No collecting data in advance “just in case”.

  8. Design principle

    Security

    Isolation per client, technology providers declared before we start, and where data is processed agreed project by project.

What we do not do

The limits, stated before signing.

A vendor who says they can do everything is a vendor to distrust. We put these limits in writing before we start, not in a footnote.

  • We do not set prices for you

    DOLMIR Software does not calculate prices or prepare quotes today. Whoever prepares the quote finds the request already read and checked.

  • We do not read drawings to calculate machining times

    It is a hard problem and we do not promise it. We automate the part that can be automated well.

  • We send nothing to your customers

    The system prepares a reply. Sending it remains a human act.

  • We do not replace your management system

    We work above what you have. If an integration is not reliable, we say so beforehand.

  • We do not use your data for anything else

    Your customers’ drawings are your customers’ intellectual property. We train nothing on them, share nothing, reuse nothing.

  • We do not promise tax incentives

    A software project does not, as a rule, qualify on its own for the Italian incentives on capital goods. Anyone who promises you it will has not read the requirements.

Data and security

Principles, not certifications we do not hold.

DOLMIR is a young company and does not claim certifications it does not hold. It states how it intends to work with data, and puts it in writing before a project starts.

  1. Minimisation

    We process only the data the agreed process needs. No collecting in advance “just in case”.

  2. Isolation per client

    One company’s customer records, catalogues and documents are never pooled with other clients’.

  3. Declared providers

    Every technology provider involved is declared before we start, with its role in processing the data.

  4. Where the data is

    Where data is processed is agreed and written down for each project. We do not take it for granted on this page.

  5. Human oversight

    No automated decision about the end customer: decisions stay with authorised people. Human oversight is part of how we work, not an option.

  6. Log of operations

    Every piece of processing must be traced and verifiable afterwards: it is a design requirement of every system we deliver.

Tell us the problem