- Home
- The system
- Swiss LLM engine
Swiss LLM engine
Fiduciary knowledge available immediately, in a private environment.
A private professional assistant specialised in Swiss fiduciary practice. It answers, researches, explains, cites, compares, asks for missing information, proposes checks, documents and escalates.
The engine researches and explains. The professional decides.
Areas covered
Specialised, not generalist.
- Swiss accounting
- Tax
- VAT
- Payroll
- Social insurance
- Financial management
- Management control
- Cost accounting
- Everyday questions of fiduciary practice
Use cases
The questions professionals actually face.
- Private-use share on a vehicle
- Provision
- VAT
- Tax treatment
- Benefit in kind
- Deductibility
- Complex entry
- Year-end review
- Inter-cantonal taxation
- Financial management
Shape of an answer
An answer that belongs in a file, not a paragraph.
Every answer follows the same format, so it can be reviewed, documented and filed.
Conclusion
Conditions
Facts to confirm
Legal basis
Accounting treatment
Tax treatment
VAT treatment
Documents
Checks
Confidence level
Date
Escalation to a human specialist where needed
Evaluation
What we measure, and how.
In internal tests based on Swiss federal examination questions in accounting, taxation and financial management, our engine reached 99 % correct answers under our evaluation protocol.
Measurement pending publication
The methodology page must publish:
- Number of questions
- Examinations used
- Years
- Subjects
- Marking scheme
- Assessors
- Model version
- Document access permitted during the test
- Results by subject
- Errors found
- Limitations
- Date of the test
Governance
An engine with guardrails.
- Validated sources
- Versioning
- Dating
- Citations
- Confidence level
- Deliberate refusal when a question falls outside scope
- Escalation
- Error log
- Correction
- Audit
- Human validation
Limitations
What the engine does not do.
- It does not replace the professional
- It makes no automatic decisions
- It requires human validation
- It depends on its sources being up to date
- It escalates to a specialist when a question exceeds it
Position on AI
What we say, and what we will not say.
Wording we use
- Production driven by controlled rules
- No public AI in client files
- Private engine
- Swiss hosting
- The professional is responsible
- Conversations not used to train an external model
- Human validation
Wording we reject
- “Zero AI”
- “Infallible AI”
- “Automatic decisions without oversight”
- “100 % Swiss” without full proof
- “99 % accurate in every situation”
Read next
-
Quality and sovereignty
Network label, standards, audits, Swiss hosting, AI policy, continuity, mediation and data portability.
-
Our evidence
Every published figure carries its date, scope, source and method. Productivity evidence file and evaluation protocol for the knowledge engine.
-
Academy
Initial training, certification, tiered progression, authorisations, continuing education and access to the network’s human centre of expertise.
You don’t need to already own a firm. You need the potential to run one.
Let’s look at whether our infrastructure can turn your project into a real accounting and trust business.