How AI works in Recavora.
Recavora uses AI in three areas: document extraction, imaging classification, and continuity intelligence. In all cases, AI output is a draft. You review and approve before anything is saved.
Philosophy
AI as your assistant. Not your decision-maker.
Medical continuity requires accuracy. That's why Recavora draws a clear line between what AI drafts and what you confirm. AI reads documents so you don't have to transcribe them. It classifies imaging so you don't have to identify study types manually. It synthesizes your history into a narrative so a new provider can understand your context quickly. But AI never saves anything you haven't reviewed, never alters a record without your confirmation, and never makes clinical judgments.
Transparency
What AI does. What AI does not do.
What AI does
- Reads and extracts structured data from uploaded documents (dates, providers, diagnoses, medications, results)
- Classifies imaging studies by type, modality, and body region
- Generates continuity stories from existing records
- Surfaces stage-appropriate continuity insights on the dashboard
- Suggests categorization and bundle organization
What AI does not do
- Diagnose medical conditions
- Make clinical recommendations
- Save any output without your explicit review and approval
- Access records from providers without patient-authorized disclosure
- Override your corrections
Commitments
Five commitments we keep
- 01No AI output is saved without your review
- 02Manual corrections are tracked
- 03AI does not diagnose, treat, or advise
- 04All pipelines are monitored for success rate, correction rate, and latency
- 05You can correct any AI-extracted field
Operations
How AI tasks are run and monitored
AI tasks are performed by dedicated edge functions, scoped to the specific record or study you initiated the task on. Other patient profiles, unrelated records, and account context are not sent to the model for an extraction or classification task. Each pipeline is monitored independently for success rate, manual correction rate, and latency. Records are processed for the task you initiated and are not used to train external models through Recavora.
Limitations
AI is not perfect — your confirmation is required for a reason
Models can be wrong. They can miss context, misread handwriting, mislabel a study, or extract a date incorrectly. The review step exists precisely because of this. Important fields should be verified before saving or sharing. Treat AI output as a starting point — never as a final answer.
Shared commitment
AI assists. You decide.
Person-directed intelligence
Every AI feature in Recavora exists to reduce friction — not to replace your judgment.
Your corrections are tracked. Your approvals are required. Your continuity stays yours.
Full overview of how AI supports continuity → /continuity-intelligence
Related
Adjacent protections
Access still requires explicit consent
AI features are bounded by Recavora's overall access model. AI does not see records it isn't authorized to see, and it doesn't create new access. Read more in Security & Data Protection.