Source before interpretation
A fact retains its provenance, date, units and version.
BIOS technology
BIOS Intelligence analyses medical data over time.
The BIOS algorithm compares facts, identifies changes and develops AI hypotheses with traceable sources, explicit gaps and limitations.
A verifiable resultPrecise factsSource · date · units
Data for the taskOnly relevant, authorised facts
The basis of a hypothesisFacts · rules · limitations
Review and controlChecks · doctor · access
The input can be reproducedThe output can be verified
What the algorithm does
It does not try to guess a diagnosis from a single measurement.
BIOS works with a fixed set of facts. Each result retains its input data, the rules applied, limitations and review status.
- Input
- Documents, lab values, symptoms and Health Memory events.
- Analysis
- Comparison over time, rule checking and the search for possible relationships between facts.
- Output
- A hypothesis to test, its basis, missing data and limitations.
- Boundary
- An unreviewed system response does not become a medical decision or a message to a patient.
Engineering principles
Facts first. Interpretation follows.
History is not overwritten
A correction creates a new version.
Gaps are part of the result
Unknowns are not hidden behind a confidence score.
Computation is separate from publication
An AI response, a hypothesis and a decision are distinct objects.
Architecture
From a document to a result with traceable provenance
Each stage creates a separate, verifiable object. Errors do not automatically pass to the next stage.
Data intake
Isolated upload
File validation
Original preserved
Common format
Type identification
Normalisation
Provenance
Health Memory
Timeline
Fact versions
Current view
Analysis
Fact selection
Longitudinal analysis
Authorised knowledge sources
AI hypothesis
Checks
Result version
Result
Professional review
Recipient-specific presentation
Audit log and monitoring
Health Memory
A versioned history of medical facts and events.
Each fact is stored with its source and time. Corrections create new versions, so the current view can be reconstructed and checked.
1
Original fact
StoredA value is stored with its source and time.
2
New version
TraceableA correction adds to the history instead of erasing the past.
3
Current view
UpdatedThe view is assembled from specific versions of the data.
4
Access
Rule-based accessRecipients see only the information they are authorised to access.
Controlled analysis
The algorithm analyses only relevant, authorised data
This makes it possible to reproduce an analysis, check sources and understand why a particular hypothesis was generated.
Selects facts
Records the exact data versions used for a particular task.
The input set cannot be changed retrospectively.
Analyses changes over time
Looks for changes, recurring patterns and possible relationships.
The source of every fact is retained.
Develops an AI hypothesis
Shows a possible relationship, alternatives, gaps and limitations.
This is a candidate for review, not a diagnosis.
Checks the boundaries
Checks structure, units, contradictions and prohibited claims.
Unresolved uncertainty stops the analysis.
Explains the result
AI chat shows the basis and review status; the clinical decision remains with the doctor.
Access to the result is checked separately.
What the algorithm provides
More than a single answer.
Four parts to every result.
Doctors and patients can distinguish medical facts from hypotheses and see what needs to be checked.
Facts
Which data was used and where it came from.
Possible relationship
What the system proposes for investigation, not as a diagnosis.
Gaps
Which data is missing for a better-supported conclusion.
Status
Whether professional review is required and who can access the result.
Safety
The system must be able to stop
An error, an unknown rule or insufficient data stops automated processing.
Unreliable input is not a fact
The document is checked or rejected.
An inactive rule is not applied
Clinical use requires separate approval.
Unreviewed AI responses are not published
Content, consent and access rights are checked.
What already works
Working code is only the beginning
Before launch, a feature undergoes technical, clinical and legal review.
A unified data flow
Versioning, provenance, event processing and Health Memory reconstruction.
Controlled analysis
Supporting evidence, structured output, medical review and safety constraints.
Clinical use cases
Bounded tasks with a responsible doctor and agreed outcome measures.
Readiness criteria
Technical, clinical, legal and independent review of the selected use case.
An AI hypothesis must be verifiable.
BIOS shows the facts used, limitations and review status. The clinical decision is made by a doctor.