Methodology
How PetCaseFinder translates veterinary research.
Every case page on PetCaseFinder is grounded in a real, peer-reviewed veterinary paper. The original authors, journal, and citation are shown prominently and link out to the source. What PetCaseFinder adds is a plain-English layer on top: a translated headline and a short summary that uses the words a worried pet owner would actually type into Google.
This page documents exactly how that pipeline works, what we preserve, and where the limits are.
1. Where the cases come from
We ingest peer-reviewed veterinary papers from five public scientific databases:
- PubMed (operated by the U.S. National Library of Medicine)
- Europe PMC
- Crossref
- Semantic Scholar
- DOAJ (Directory of Open Access Journals)
For each paper we store the original title, original abstract, authors, journal, year, country, institution (when available), and a stable identifier that links back to the source database. We never modify the original metadata.
2. Filtering for veterinary relevance
Not every paper that appears in the source databases is relevant to pet owners. Before we touch the translation step we run a multi-signal vet-relevance filter that excludes:
- Human-only clinical research
- Materials science, engineering, and lab-instrument papers
- Lab-rodent-only studies with no companion-animal angle
- Papers with no animal mention in the title or abstract and no veterinary signal in the journal name
Filtered papers are kept in the database (so we do not redo ingestion later) but are not surfaced in the case index and are not translated.
3. The plain-English translation
For each remaining paper we generate two things using OpenAI's language models:
- A plain-English headline. A single owner-facing sentence that captures the species, the symptom, and the outcome. This becomes the page's H1 and the <title> tag.
- A plain-English summary. A 3 to 5 sentence narrative that opens with what happened to the animal, walks through what the vet tried, and ends with the outcome. This becomes the meta description and the highlighted summary box on the page.
The model is prompted with strict, non-negotiable rules. It must stay faithful to the source, must never invent symptoms or treatments, must not introduce people or animals that the abstract does not describe, and must use plain words like "vomiting", "limping", "tumor" instead of the clinical terms. If the model cannot produce a faithful, specific headline from the abstract, it returns a sentinel value and the page keeps the original academic title instead.
4. The original abstract is always shown
The plain-English summary never replaces the original abstract. Every case page renders both: a translated summary at the top and the verbatim, unmodified abstract directly below it. The original authors, journal, year, and institution are shown in the byline under the headline. The original publication title is shown verbatim under the rewritten one. A direct link to the original paper on PubMed, Europe PMC, or via DOI is shown in the byline and again at the bottom of every case page.
5. What we preserve
- The original paper title
- The original abstract, verbatim
- The full author list
- The journal name
- The publication year
- The author institution and country, when available
- A working link to the paper on its source database (PubMed, Europe PMC, Semantic Scholar, or DOI)
- The DOI, when the source provides one
6. What we will never do
- We will never invent symptoms, treatments, doses, durations, or outcomes that are not stated in the original paper.
- We will never recommend a treatment to a pet owner. The translated summary describes what the original paper reported, not what the reader should do.
- We will never strip or hide the original authors, the journal, or the citation.
- We will never publish a page sourced from a paper that is not indexed in a public peer-reviewed scientific database.
7. Limits and how we monitor them
Language-model translation is not perfect. Edge cases exist where the rewritten headline can be vague or where a clinical term gets simplified more than ideal. We continuously sample translated pages, audit them against their source abstracts, and reset individual rows for re-translation when a quality regression is detected. The original abstract on every page is your safeguard: if anything in the plain-English layer reads off, the verbatim source is a single scroll away.
If you find a translation that misrepresents the source, please email hello@petcasefinder.com with the URL — we will reset and re-translate the page.