How it works

Spokojnie answers the questions Polish parents of children aged 0–3 actually type into search engines: is this normal at this age, when should I worry, and where should I go. Every sentence of an answer is a claim backed by a verbatim quote from a trusted source. Nothing is written from memory.

1. Who and what is behind it

The site is run by Piotr Kozłowski. Answers are produced by a detailed process built on AI models from Anthropic (Claude). We say so plainly, because with a child's health it matters where an answer comes from.

AI reads the sources, including ones in other languages (for example the UK's NHS or WHO), summarizes them in Polish and drafts the answer. Quotes stay in the original language so you can check them.
Separate models check the draft: they open every source and compare the quote with the claim.
AI does not add anything from memory, does not assess any individual child and does not decide what is published. A person decides, and Piotr Kozłowski is responsible for the content.

2. Start from real questions

Search autocomplete phrases from Polish parents were collected and grouped into canonical questions with an age window. The site covers the questions parents ask most often, not the ones that are easiest to write about.

3. Build each answer from sources only

For each question, a research agent drafts a record using only sources from a fixed registry of 22 public-health institutions and publications. Each claim carries a link and a verbatim quote of at most one sentence. If no registered source covers something, the record says so in a gap list instead of filling it in.

4. Check every claim twice

A separate fact-checking agent, which never sees the author's reasoning, opens every source and confirms that each quote is verbatim and supports its claim. Records about symptoms get a second, independent check on a different model. Only claims confirmed by both checks can appear on the site.

5. A person approves every batch

A person reviews and approves records in batches. Any edit after approval invalidates it until the record is approved again.

6. Safeguards instead of a clinical reviewer

The project has no clinical reviewer, so four safeguards are enforced in code before anything is published:

S1 — only confirmed claims are shown.
S2 — symptom records need a second check on another model.
S3 — every page shows its verification date and a "Report an error" link.
S4 — some topics stay off the public site: parents' own mental health, and emergencies, except narrow pages that a Polish source covers and that passed both checks. There is no medicine dosing anywhere.

Emergency signs come first: the "ER / 112" panel always appears above "See a doctor."

7. Gates before every deploy

A deploy runs only through one script. It stops if the test suite, the record validator, the S1–S4 deploy gate or the content invariants fail. Quotes are rechecked against their source pages every month; a quote that disappears takes its record off the site.

8. This English edition

The English pages are translations of approved Polish records, made by one agent and checked field by field by another on a different model. Quotes stay in their original language. A translation is published only while it matches the current approved Polish record. The care paths (112, night and holiday care, primary care) are Polish.

In numbers

278 approved answers, 123 in English
3,414 sourced claims
17 sources cited
170 symptom records with a second check

Spokojnie is a non-commercial project: no ads, no affiliate links, no paid features.