Intelligence architecture

How biomarkr thinks about your blood results.

biomarkr does not simply ask AI to read a PDF. It structures your results first, compares them over time, connects related biomarkers across body systems, and checks generated explanations before they reach you.

Facts first. AI second. Always non-diagnostic.
See a sample report → Back to homepage
Structured biomarker output · example
Your iron levels · Ferritin
28 µg/L
↓ Trending down
Body system
Iron & oxygen transport
Reference context
30–150 µg/L
Direction
↓ Falling · 56% from baseline
History
5 tests · 18 months
Validated output
"Ferritin has fallen across four consecutive tests. This pattern may be worth discussing at your next GP appointment."
6-stage intelligence pipeline · validated · non-diagnostic
Core principle

Facts first.
AI second.

Most blood test tools start with a document. biomarkr starts with structured data.

A generic AI given a PDF can only work with what the document says. biomarkr extracts, validates and enriches biomarker data before AI sees any of it — so the explanation is built on facts, not on text.

Before any explanation is written, biomarkr separates out the biomarker name, value, unit, reference range, test date, previous result, direction of change, and body-system context. Then AI helps explain what those facts may mean — within clear safety boundaries.

Most tools
PDF → AI reads text → Answer
biomarkr
PDF result → Structured data → Trends → Body systems → Explanation → Validation
Intelligence architecture

The biomarkr intelligence pipeline,
step by step.

biomarkr runs a multi-stage intelligence pipeline on every uploaded result — structuring biomarkers, scoring body systems, comparing history, reasoning across connected markers, grounding explanations in curated context, and checking outputs before they reach you.

Structured intelligence first. Controlled AI explanation second.
01 · Ingest
Extract
OCR · PDF · image
structured input
partner / API data
supported UK lab formats
candidate extraction
02 · Structure
Validate & score
name validation
unit normalisation
reference context
body-system scoring
9 systems · weighted
03 · Reason
Multi-marker intelligence
longitudinal trends
conflict detection
ferritin↑ + CRP↑
→ inflammation context
multi-marker reasoning
04 · Ground
Evidence-aware context
curated guidance context
marker-specific evidence
allowed claims
forbidden claims
fact pack assembled
05 · Explain
Controlled AI explanation
controlled AI explanation
constrained by fact pack
narrative controls
trend language level
severity framing
06 · Validate
Safety & source checks
narrative linter
safety validator
source-data check
numeric claim check
deterministic repair
01 · Ingest
Extracts structured biomarker data from supported UK lab formats.
PDF · image · structured input · partner API
candidate biomarker extraction · supported UK lab formats
02 · Structure & Score
Validates, normalises and scores across nine body systems.
Name validation · unit normalisation · reference context
weighted body-system scoring · Iron, Thyroid, Liver, Kidneys, Lipids, Metabolic, Inflammation, Hormones, Nutrients
03 · Multi-Marker Reasoning
Reads connected markers together and resolves conflicts before any explanation is written.
Longitudinal trend analysis · narrative mode detection
Reasoning examples:
Ferritin↑ + CRP↑ → inflammation context flagged
Glucose normal + HbA1c↑ → different time-window signals
B12 normal + MMA↑ → functional mismatch, uncertainty noted
Resolved into cautionary directives before explanation
Explanation layer begins here
04 · Evidence-Aware Grounding
Grounds the explanation in curated clinical context drawn from published NHS and NICE guidance.
Marker-specific evidence context · curated guidance
Scoped to markers present in this report
Fact pack assembled: allowed claims · forbidden claims
Narrative controls: system ranking · trend language level · severity framing
05 · Controlled AI Explanation
AI generates the explanation — constrained by the fact pack, narrative controls and non-diagnostic boundaries.
Controlled AI explanation · constrained by assembled fact pack
Narrative controls assign: headline system · trend language level · severity tier
Generates: home narrative · biomarker narratives · system summaries · next steps
Narrative mode: baseline · early_trend · longitudinal
06 · Safety & Source Checks
Three independent validation layers check the output before it reaches you.
✓ Narrative linter — unsafe wording · trend-language compliance · banned phrases
✓ Safety validator — directional contradictions · claim/fact mismatches · deterministic repair
✓ Source-data check — every numeric claim verified against source biomarker values
✓ No additional AI call — repair is template-based and deterministic
Intelligence pipeline — processing in real time
Stage 01 · Ingest & Extract
 
Validated output · source-data checked · non-diagnostic boundaries confirmed
Connected reasoning

Biomarkers rarely make sense alone.

A single marker can be useful. A connected pattern is often clearer. biomarkr reads related markers together within each body system.

Iron & oxygen transport
Ferritin, haemoglobin, transferrin saturation and iron markers are read together to help explain the broader pattern — because each reflects a different aspect of iron status and oxygen carrying capacity.
Ferritin Haemoglobin Transferrin sat. Iron RBC
Metabolic health
Glucose, HbA1c, insulin and triglycerides are considered together — because they reflect different time windows of metabolic activity, from a single moment to a three-month average.
Glucose HbA1c Insulin Triglycerides
Inflammation context
CRP and ESR can change how other markers, such as ferritin, should be read — because ferritin is an acute-phase protein that rises with inflammation, not just low iron.
CRP ESR IL-6 Ferritin (context)
biomarkr helps explain patterns. It does not diagnose conditions.
History matters

Your history changes
the interpretation.

One test can show where you are today. Repeat tests can show where you are heading.

Show whether a marker is rising, falling or stable across tests.
Compare current results with previous results — not just a reference range.
Avoid claiming a trend when there is no historical data to support one.
Build a longer record as more results are added over time.
Help distinguish a one-off result from a repeated pattern of movement.
The example report shows 18 months. biomarkr is designed to build a longer longitudinal record as more results are added.
Your iron levels · Ferritin
Trending down
28 µg/L ↓ from 64
64
Sep '24
52
Feb '25
41
Jul '25
36
Dec '25
28
Apr '26
Five results across 18 months. A single test would not show this pattern. Each new result extends the record.
Evidence-aware

Guidance-aware,
not generic AI alone.

biomarkr combines structured user data with curated clinical guidance and evidence-aware context where relevant — helping explanations stay more consistent and cautious than generic AI responses alone.

What a marker is commonly used for — so explanations have a factual basis, not just a value.
When related markers matter — so connected biomarkers are considered where they add context.
When clinician discussion may be appropriate — so biomarkr points toward a professional rather than away from one.
Where uncertainty should be stated clearly — so biomarkr does not overclaim when the data is limited or the result is ambiguous.
Guidance does not override your actual data. biomarkr does not use it to diagnose, and does not imply that every output is clinically reviewed or validated against any specific set of guidelines.
Safety & boundaries

The model can explain.
It cannot invent your facts.

Before explanations reach the user, biomarkr checks for unsafe or unsupported claims. If biomarkr does not have enough information, it should say so. A missing result is not a licence to guess.

"A pattern of consistent decline across four tests is meaningful. A single result below range may not be. biomarkr tries to reflect that difference."
No invented values. Explanations are grounded in what the data actually shows.
No unsupported trends. A trend is only stated when there is history to support it.
No diagnosis. biomarkr explains patterns. It does not name conditions or causes.
No treatment instructions. biomarkr does not prescribe, recommend medication or advise starting or stopping treatment.
No hidden uncertainty. Where data is limited or ambiguous, biomarkr states that clearly.
Clinician escalation flagged. Anything that may need professional review is noted — not left to the user to infer.
No causality claimed from interventions. biomarkr cannot confirm what caused a result to change.
One marker is not the whole story. Results are read in body-system context, not in isolation.
One intelligence layer

Multiple surfaces.

The goal is not to automate clinical judgement. It is to make blood test interpretation clearer, more consistent and easier to review.

The same structured intelligence model supports patient experiences, clinician-reviewable reports and partner outputs — without compromising on claim discipline or safety boundaries.

Patient app
For patients
Clear explanations, system scores, trends and next steps — written for someone reading their own results, not a clinician reviewing them.
For clinicians
Clinical reports
Structured, patient-friendly reports designed for clinician review — with source data, trends and flagged markers included for context.
For partners & labs
Partner / API
Structured interpretation outputs — including summaries, system scores, trend data and next-step guidance — for health platforms, labs and testing providers.
See it in practice

See the intelligence
in a real report.

The best way to understand biomarkr is to see how it explains one blood test over time — structured, connected and validated before it reaches you.

See a sample report → Join early access