ArkeoAI responds like an experienced colleague who knows your documents
Get a fast, personalized answer to every question.
A larger model does not mean a better answer!
Contrary to popular belief, the quality of AI responses is not primarily determined by the size of the model. Research shows that 40–50% of response accuracy comes from data quality — that is, what documents, policies, and professional materials have been loaded into the system. A further 25–30% comes from prompt quality: how is the question asked? The model itself is responsible for only 20–22% of the result. This means that a well-prepared, targeted system with a smaller model delivers more reliable answers than a large general-purpose AI with imprecise data and poorly formulated questions. To get the most out of the system, using good prompts is essential.
One of the most important factors in using artificial intelligence is how you formulate the question. The accuracy of the answer can be seen across several phases.
Using good prompts is essential to getting the most out of the system
One of the most important factors when using artificial intelligence is how you phrase the question you ask. The accuracy of the response can be observed across several phases.
Topic: employment law — statutory notice period calculation (French Labour Code / Code du travail). The correct answer is short and objective — precision matters, not length.
Level 1 : Weak
« What is the notice period? »
AI answer: Notice periods vary depending on the situation. It is generally one to three months, but you should check the applicable rules.
Why it fails: The prompt provides no context at all — no contract type, no seniority, no party initiating termination. The AI has no choice but to give a vague, unusable answer. Even if technically harmless, it offers no actionable value to a professional.
Level 2 : Fair
« An employee has worked at a company for 6 years. What is their notice period? »
AI answer: Under French labour law, for 6 years of seniority the statutory notice period is 2 months.
Why only fair: Seniority is given, but who is terminating? Under French law (Art. L1234-1 Code du travail), the 2-month period applies to employer-initiated dismissal for employees with 2+ years’ seniority. For employee resignation, notice is set by collective agreement (convention collective) — often different. The answer is incomplete and may mislead.
Level 3 : Good
« An employer is dismissing an employee with 6 years of seniority (non-managerial, indefinite contract). What is the statutory minimum notice period under Art. L1234-1 of the Code du travail? »
AI answer: Under Art. L1234-1 Code du travail: employees with 6 months to 2 years’ seniority: 1 month; over 2 years: 2 months. For 6 years of seniority, the statutory minimum is 2 months.
Why good but not optimal: The legal reference is correct and the answer is accurate. However, it does not flag that a collective agreement (convention collective) may provide a longer notice period — which always prevails over the statutory minimum. A practitioner needs that caveat to avoid advising incorrectly.
Level 4 : Optimal
« An employer is dismissing a 50-year-old employee with 6 years of seniority (non-managerial, indefinite-term contract, no gross misconduct). What is the statutory notice period under Art. L1234-1 Code du travail? Show the calculation steps and note any factors that could modify the result in practice. »
AI answer: Art. L1234-1 Code du travail: seniority > 2 years → statutory minimum = 2 months.
Age: no statutory age modifier under French law (unlike some other systems).
Practical modifiers to check:
— Collective agreement (convention collective): may extend the period (always check).
— Employment contract: may stipulate longer notice; cannot be shorter than the statutory floor.
— Gross misconduct (faute grave/lourde): would reduce notice to zero — not applicable here.
Conclusion: minimum 2 months, subject to applicable convention collective.
Why it’s optimal: All relevant parameters are specified up front (age, seniority, contract type, no misconduct). The answer gives the statutory floor, explains what can modify it, flags the collective agreement caveat, and states what does not apply and why. A lawyer or HR professional can act on this immediately — and knows exactly what to verify next.
How the system works — in three steps
No expertise is required; a single person can complete every step in minutes.
You upload your own documents
The system retrieves the relevant data based on those documents
You receive a precise answer to your question (answers can be downloaded in .rtf format)
Why do many AI projects fail? The reason is: lack of data preparation!
Many companies underestimate the prerequisites for deploying an AI model: raw documents (DOCX, PDF, paper) cannot be used directly to build a proprietary database (RAG). Complex data preparation is required: text extraction, cleaning, RAG indexing. The problem is especially pronounced with paper-based documents: OCR + manual verification — the risk of data loss can be significant. This service is typically provided by specialist companies, usually at €0.10–0.70 per page (generated documents and standard PDFs are easier to prepare; paper and scanned PDFs are considerably more difficult).
We process your confidential paper documents on-site in a uniquely secure way: scanning on your premises, OCR + RAG indexing. Your data does not leave the room for a single moment. Within our service, we process 1,000 pages free of charge.
What our ArkeoAI guarantees
No cloud
Your data stays on-site
No internet
Zero connection required
Local model
AI runs in your office