Complete transparency

What actually happens to your medical documents?

Many people hesitate to send their medical records to an online service. The questions are natural. This page answers all of them, plainly.

See consultation types

“Does AI make mistakes? Does it invent medical information?”

This is the most frequently asked question. And it’s a good one. Generic AI models have a real problem: they invent information with an air of authority. Doctors have a term for it: hallucination.

An ordinary AI responds from everything it has read on the internet. Some sources are good, others are wrong, and the model doesn’t always tell them apart.

Why OncoExpertAI works differently

The system does not respond from its general memory. It responds exclusively from the medical knowledge base built and validated by Dr. Onisim Florin, comprising over 230,000 pages of international protocols and studies.

Technically speaking, it’s like asking a doctor who is only allowed to consult verified medical books — not internet forums. It cannot invent something that doesn’t exist in those books.

And before you receive the report, Dr. Onisim Florin reads it personally and validates it. No document reaches you without having passed through a human medical eye.

You send the documents. Here’s what happens next, step by step.

The six steps below describe exactly what the system does with your information, from the moment you upload the documents until you receive the report.

What information is the analysis based on?

The short answer: a medical knowledge base built over the past years by Dr. Onisim Florin, specifically designed to be used by artificial intelligence. Not data taken from the internet, not press articles. International protocols, clinical studies, treatment guidelines.

To give you a clearer picture: 230,000 pages means several hundred medical textbooks. No doctor can read them all, or keep them all in memory. The system can.

Where does the information come from?

The knowledge base is aligned with the most important international medical sources:

  • ESMO
  • ASCO
  • NCCN
  • AJCC
  • PubMed (40M+ articles)
  • UpToDate
  • ClinicalTrials.gov

How does the system prevent it from inventing information?

This question deserves a clear technical answer, translated into plain language.

The technique is called RAG, short for Retrieval-Augmented Generation. In plain terms: the AI model does not respond from what it knows, but from what is placed in front of it at each session.

Imagine a doctor who, before each consultation, receives a stack of files specific to your case. They are only allowed to respond based on what they find in those files. Not from what they remember from ten years ago, not from something they read somewhere. Only the files on the table.

The OncoExpertAI system works exactly like this. At each session, the modules relevant to your case are loaded into context, and the AI model cannot generate responses based on anything else. Technically, it is constrained to respond exclusively from the validated information activated for that session.

It understands intent, not just words

When you describe a symptom in your own everyday words, the system does not search for that exact phrase in the database. It uses vector representations of text, a technique that allows it to understand the meaning and find the right module even when your terminology differs completely from formal medical language.

If you say “I have pain in the stomach area and I’ve lost a lot of weight”, the system can correlate this with the relevant clinical parameters without you needing to know the exact medical terms.

Which AI models are used and why three, not one?

No AI model is perfect in every situation. Some are better at processing long documents, others at nuanced and empathetic responses. The solution was not to choose one, but three — each used where it excels.

Google Gemini 2.5 Pro

Processing long medical documents, maintaining coherence across dozens of pages.

Google Gemini 3.1 Pro

Understanding complex structures in files with multiple investigations and varied document types.

Anthropic Claude Sonnet

Formulating responses: clear, empathetic, adapted for a patient, not for another doctor.

The result: a report that has the precision of a medical document and the readability of a text written for a person, not a specialist.

And what if the documents are in Romanian while the modules are in English. How does that work?

All medical modules are written in English, the language in which the most extensive specialist literature exists. The system has integrated automatic translation, and the dialogue with the patient takes place in any language, without the medical precision being affected by translation.

Why does it take 3 to 48 hours, if AI is fast?

The AI generates the analysis in a matter of minutes. But the report does not go out immediately.

Dr. Onisim Florin reads every report before delivery

The 3 to 48-hour window exists because every generated document goes through a human medical review. Dr. Onisim Florin reviews it, supplements it if needed, and personally approves each report.

This is not a formality. It is the guarantee that the speed of technology does not come at the expense of safety. You receive an analysis that has been considered both by an algorithm and by a doctor with real clinical experience.

Processing long medical documents, maintaining coherence across dozens of pages.

This is the approach known as “Human-in-the-Loop” — the human remains in the final decision loop. The system does the heavy work of synthesis. The doctor makes the final validation decision.

What happens to your medical documents?

Medical data is among the most sensitive information you have. Questions about confidentiality are not paranoia — they are common sense.

  • All data is encrypted in transit via SSL connection
  • Documents are stored in systems with strictly controlled access
  • The platform fully complies with GDPR regulations
  • You can use the service completely anonymously, without disclosing your identity

The anonymity option is a real one, not a footnote detail. Remove your personal data from the documents, provide a first name or pseudonym, your real age and biological sex, and you receive an equally complete analysis. Without any trace of your identity.

What does the report you receive actually contain?

Not a two-page paragraph. Not generic recommendations along the lines of “consult a specialist”.

The report can run to 20+ pages and covers:

All treatment options currently available worldwide

Not just what is done in Romania, but what exists globally, with guidance on whether and how you can access those options.

How cancer interacts with other conditions you have

If you also have diabetes, hypertension or other conditions, the analysis takes that into account and indicates the necessary adjustments.

What may happen in the future and what to do in each scenario

Possible complications, recurrence, metastasis. Not to frighten you, but so you are not caught unprepared.

Answers to the questions you submitted

You send your questions at the time of ordering. The report includes them and answers each one explicitly.

The report does not replace your treating physician. It is a tool that you or your doctor can use as a reference point when making therapeutic decisions.

Now you have the answers. The next step is yours.

Choose the type of analysis that fits your situation and send the documents. You will receive the report within 3 to 48 hours.

See available consultations