Documents
A document parser first extracts text, tables, and document signals. The AI Worker then asks a configured model for schema-constrained output and validates the response before callbacks are accepted.
AI methodology
How Hoom uses AI while preserving source evidence
Hoom uses AI as an organization and reasoning layer over a patient-controlled archive. The AI does not replace the source record or a clinician, and different operations can use different configured model providers.
Possible consideration
Evidence-linked review item
Why this may be relevant based on recorded facts.
- Supporting evidence with source
- Conflicting evidence
- Missing information
Designed for continuity
Keep the useful detail without losing the larger story
Notes
User text can be sent to a routine model to identify structured medications and symptoms. Users can instead enter those facts manually without AI processing.
Summary
A summary organizes available health history into Markdown. It is informational and can be refreshed when source data changes.
Health Review
A separately configured review model evaluates facts and must return structured evidence, conflicting evidence, missing information, and follow-up questions.
Known limits
Models can omit facts, invent unsupported details, misread units, or reason incorrectly. Validation reduces malformed output but cannot prove medical correctness.
Your history stays connected
Start a patient-controlled archive
Bring records, facts, and time together before the next appointment.