How bloodwork parsing works

Effective date: 2026-05-13

Why this page exists

Peakdose can pull the values out of a bloodwork report for you instead of asking you to type each number by hand. This page explains, in plain language, what happens to that file, what gets sent to an AI service, and what stays out of it. Using parsing is your choice, and you can change your mind at any time.

What happens when you upload a report

  1. You choose a PDF or image of your results and confirm the upload.
  2. The file is saved to your private storage in Supabase, hosted in London. Only your account can read it.
  3. A short Peakdose function passes the file, along with our internal list of marker names, to an AI parsing service.
  4. The service returns a structured response: a value, a unit, a lab reference range when present, and a suggested match against the markers Peakdose tracks.
  5. You see a review screen with every extracted value before anything is saved to your history. You can edit any field or discard the lot before continuing.
  6. When you confirm, the values are saved to your bloodwork history alongside the file.

What the AI service sees

It receives the file bytes you uploaded, a parsing prompt written by Peakdose, and our catalogue of marker names. That is all.

It does not receive your name, your email, your account identifier, your dosing history, your other health data, or anything else stored in Peakdose. The service has no view of who you are.

What the AI service returns

It returns numbers, units, reference ranges, and suggested marker matches. It does not return diagnosis, interpretation, or advice. Peakdose does not show interpretation either; we show your trends, and you decide what to do with them.

Parsing is usually accurate on clearly formatted reports. It can misread unusual layouts, smudged scans, or unfamiliar units, which is why every value passes through your review screen before it is saved.

Where the AI service runs

Parsing uses AWS Bedrock with Anthropic's Claude Sonnet 4.6 model, running on the EU cross-region inference profile. Requests are routed across Bedrock endpoints in Ireland, the United Kingdom, Germany, and France. Your data does not leave the European Economic Area.

AWS (Amazon Web Services EMEA SARL) is our named sub-processor. The relationship is bound by a Data Processing Agreement that contractually excludes your data from model training and limits retention to the duration of the parse call. Anthropic provides the model but does not receive your data when it runs on Bedrock.

Your control

You never have to use parsing. You can decline at the consent prompt and enter values by hand. If you have used it before, you can revoke consent at any time under Settings → Bloodwork parsing. Deleting a bloodwork entry removes both the values and the underlying file; deleting your account removes everything.

More detail

For how Peakdose handles your data more generally, see our Privacy Policy. For questions, contact us at contact@peakdoseapp.com.

Peakdose Ltd