Skip to content
AI Medical Assistant Your Health Companion

How it works

The AI writes the words. The program makes the decisions.

That single rule shapes the whole design. Here is what happens between your question and the answer on your screen.

Asking a question, step by step

  1. 1. You type your question

    Anything in your own words, in any of the three languages.

  2. 2. The input is validated

    Length and content are checked before anything else happens.

  3. 3. Emergency rules run first

    Your words are matched against eight red-flag rules by plain program logic. If one matches, the flow stops here and you are shown emergency guidance and phone numbers. The AI is never called.

  4. 4. Your usage allowance is checked

    Each account has an hourly and a daily request limit, so the free API tier is never overrun.

  5. 5. The prompt is assembled

    Safety instructions first, then the task, then your words — clearly marked as data, not as commands the model should follow.

  6. 6. The model writes an answer

    Google Gemini receives the prompt over a server-to-server call. Your browser never touches the API or the key.

  7. 7. The answer is validated

    Empty, malformed, diagnostic, prescription and dosage-shaped responses are rejected. An answer that dismisses a detected emergency is thrown away.

  8. 8. You see the result

    The answer is saved to your conversation and shown. If anything failed, you get one clear message — not a technical error.

What the AI is told, and in what order

The order matters. Anything from you, or from a file you upload, arrives labelled as material to be analysed — never as instructions.

1

Safety instructions

Trusted

You are not a doctor. Do not diagnose. Do not prescribe. Do not give dosages. Say when you are unsure. Advise urgent care when it may be needed.

2

The task

Trusted

For example: summarise this report using only the values it contains.

3

Your question

Treated as data

Your own words, passed through untouched.

4

Uploaded document text

Treated as data

Fenced off and explicitly labelled. If a report contained the line "ignore previous instructions", it would be read as text inside a document, not obeyed.

Uploading a report

The file is checked

Type and size are both validated. Only PDF, text and image files up to 10 MB are accepted.

It is stored privately

Saved under a generated name in a folder the web server does not serve, so the original filename cannot affect where it lands.

Text is extracted

From text-based PDFs and text files. If no readable text is found, the file is kept and marked unreadable — nothing is guessed.

The text is summarised

Only the extracted text is sent. The summary is built from your document, and you can open the extracted text to check.

When something goes wrong

Failures are shown honestly rather than hidden behind a spinner or a fake success.

The AI service is unavailable

Your question is still saved. You see one plain sentence asking you to try again later — never a technical error or the name of the provider.

You reach your request limit

You are told clearly that the limit is reached, and when to come back. A failed request does not count against you.

A report cannot be read

The file is kept and marked as unreadable. The app never claims an analysis happened.

The AI writes something unsafe

The response is rejected before it reaches you, and you are told the answer could not be produced.

See it for yourself

Ask a question, then try mentioning chest pain and watch the emergency rules take over.

Create an account