How the blood-test.life AI blood test analyzer works

Every step of how your lab report becomes a doctor-grade plain-English analysis — what we extract, what the AI checks, and what the report tells you.

The six-stage AI analysis pipeline

What happens between the moment you drop your PDF and the moment you read your report.

Upload your lab report

Drag a PDF, take a phone photo, paste a scan or screenshot. Any layout. Any language. Reports from Quest, LabCorp, NHS, Synevo, Mayo Clinic Labs, Synlab, Bupa, and 400+ other labs are recognized out of the box.

The parser extracts every value

Our template library matches your lab's format and pulls every biomarker into a structured record. When the layout is unfamiliar, a vision-language model reads it like a careful intern would — and the new template is added to the library so the next user gets a faster result.

The normalizer adjusts everything

Units get converted to your locale's convention (mg/dL ↔ mmol/L). Biomarker names map to LOINC codes. Reference ranges shift for your age, sex, and pregnancy status if relevant. This is where most of the meaningful work happens — long before the AI writes anything.

The clinical-rules engine runs

Validated rules fire deterministically — diabetes thresholds, eGFR staging, lipid risk calculators, anemia classification logic. The rules engine ensures critical interpretations never depend on a language model alone.

The narrative model writes the report

A fine-tuned, domain-trained language model writes the patient-facing narrative — constrained to phrasing that has been reviewed by our medical advisory board. Every claim it makes is tied to a biomarker actually present in your report.

You read the report

On-screen first, then downloadable as a clean PDF. Three audiences in one document: plain English for you, a doctor-ready paragraph for your physician, and a machine-readable JSON for trend tracking.

What you'll see in your report

Every AI blood test analysis we produce contains six layers.

Biomarker table

Every value the lab reported, normalized, with a clear flag — normal, borderline, high, low — and the reference range we applied.

Personalized ranges

Age- and sex-adjusted at minimum. Pediatric, pregnancy, and athlete-adjusted where relevant.

Plain-English narrative

Top 3–5 findings, prioritized by clinical relevance, in the language you chose at upload.

Doctor-ready summary

A one-paragraph clinician synthesis that fits the front of a chart in under a minute.

Concrete next steps

Re-test intervals, lifestyle adjustments, specialist referrals — specific, not generic.

Audit trail

Model version, reference data sources, reviewer name, confidence score, and the medical disclaimer.

What the AI cannot do

Honesty matters more than marketing here. Three things the AI explicitly does not do:

Diagnose disease. A pattern in your lab report may strongly suggest a condition, but only your clinician can integrate it with your history, symptoms, and physical exam to make a diagnosis.
Prescribe medication. Even when supplementation is obvious (vitamin D, iron, B12), we recommend you confirm dosing with your physician — drug-drug interactions and absorption issues can't be inferred from a lab report alone.
Replace a follow-up appointment. The AI's job is to prepare you for that appointment, not to skip it.

When the AI's confidence in a result drops below threshold — faint scan, ambiguous unit notation, partially handwritten report — it tells you so explicitly rather than pretending. That's the difference between a tool and a toy.