Our approach
How AuVentures turns scattered records into a history you can trust.
A plain account of how the work actually works — what the tools do, how patient information is protected, and where human judgment stays in charge.
The problem
Complex care produces fragments, not records.
A patient with a complex neurodevelopmental or autoimmune condition may see many clinicians across many systems over many years. Their history accumulates as scanned PDFs, faxes, lab reports, and imaging results — each in a different format, none of it connected.
The result is that no one holds a complete picture. Clinicians reconstruct what they can in the minutes before an appointment. Patients and families carry the rest in their memory and their folders. The information exists; it is simply never assembled.
AuVentures is built to do that assembly — carefully, consistently, and in a way that keeps the patient in control of their own information.
How it works
1
Capture
Patients and care teams bring records into AuVentures through a simple upload — adding documents from a desktop or dropping files in directly. There is no requirement to integrate with a hospital system to begin.
2
A patient-level knowledge base
Each patient’s information is held in its own knowledge base — organized around the person, not the practice or the visit. This is the foundation that lets a history be built and kept consistent over time.
3
Digitize
Most records arrive as images and scans that software cannot read. AuVentures uses optical character recognition — Tesseract and additional processing logic — to convert these unstructured documents into structured, machine-readable data.
4
De-identify and protect
Before any data is prepared for AI reasoning, identifying information is removed. Sensitive fields — including details that are clinically informative but personally identifying, such as date of birth and gender — are deliberately withheld from any external model. Records are encrypted in storage and in transit.
5
Reason
De-identified information is passed to a commercial large language model — currently OpenAI’s API — for the reasoning layer: interpreting documents, recognizing what a record describes, and raising plausible directions worth a clinician’s attention.
6
Store, trend, and analyze
Records are kept in a relational database in cloud storage. Trending and longitudinal analysis happen here, in structured data — not inside the language model — so the patient’s timeline stays consistent and accurate rather than being re-interpreted with each pass.
Why the architecture matters
The language model reasons. The record is governed elsewhere.
This is the most important design decision in AuVentures, and it is worth stating plainly. Large language models are capable at interpretation, but they are not reliable at holding a long, precise history consistent — dates, dosages, and sequence can blur from one pass to the next.
So AuVentures uses the language model for reasoning, and keeps the patient’s actual timeline in a structured relational database where consistency can be enforced.
The model helps interpret and surface. The database holds the truth of the record. Separating those two jobs is what allows the system to be genuinely useful for complex care without inheriting the unreliability that general-purpose AI tools show when they are handed the whole task.
Privacy by design
Identifiable information stays inside protected infrastructure.
Patient privacy in AuVentures is enforced by architecture, not promised by policy. Two distinct protections do two distinct jobs.
Encryption protects records in storage and in transit, so that information cannot be read by parties who should not have it. De-identification removes the personal identifiers from data before it is sent to any external model — so that the information used for reasoning cannot be traced back to an individual.
Identifiable patient information is stored within secure cloud infrastructure. The commercial language model used for reasoning receives only de-identified data and never receives the sensitive identifying fields that are deliberately withheld from it.
Where judgment stays human
The tools inform. Clinicians decide.
AuVentures can organize a history, surface a pattern, and raise a plausible direction worth considering. It does not diagnose, and it does not decide. Determining what holds up clinically — what is valid, what matters, what to act on — remains the responsibility and the judgment of the clinician.
This is a deliberate limit, not a temporary one. The purpose of the work is to give clinicians a clearer picture and more time to think — not to substitute for the judgment, accountability, and relationship that care depends on.
Honest about the stage
What we can and cannot yet claim.
AuVentures is a pilot-stage organization. The architecture described here is how the system is built; the evidence base for its impact on care is still being developed, and we would rather say that directly than overstate it.
What we can offer — to patients, to clinicians, and to funders — is a transparent account of how the tools work, a design that puts patient control and human judgment first, and a genuine commitment to documenting what we learn, including what we get wrong.
Questions about methodology, data handling, or ethics: privacy@auventureshealth.org