×
One connected health record
the patient actually owns
Wearables, documents, and the doctor visit — reconciled into one timeline the patient owns, with Claude doing the reasoning.
A health story scattered across three places that never talk
A patient's health story is scattered across three places that never talk to each other: the continuous signal from wearables, the paper trail of lab reports and records, and the spoken doctor visit — the most decision-dense moment of all, which disappears the second it ends.
People juggle a fitness app, a folder of PDFs, and a fading memory of what their doctor told them. No single product pulled all three together, and research puts patient recall of a visit at 40 to 80% forgotten, with much of what is remembered recalled incorrectly.
DiagNova set out to be one of the first consumer apps to capture and reconcile all three into one timeline the patient owns — and to make sense of it without ever crossing into diagnosis.
Three pipelines, reasoned across as one record
We built an app that connects a patient's whole health picture through three pipelines, then reasons across them as one record. The continuous signal, the paper trail, and the conversation finally inform each other.
Wearables
Through a single aggregator integration, DiagNova connects Apple Health, Oura, Whoop, Fitbit, Garmin, and CGMs, normalizing every stream onto one timeline. Patients connect once instead of living in an app graveyard.
Health document uploads
Patients upload any lab report, discharge summary, or prescription as a PDF or photo. AWS Textract handles OCR, and Claude structures the raw text into typed biomarkers, medications, and dates — deduplicated onto the timeline, with the source page kept for citation. A drawer of paper becomes a searchable, trended history.
Visit Intelligence
Patients record the appointment on their phone with a logged consent step. AWS HealthScribe produces a speaker-labeled, evidence-linked transcript, and Claude turns it into a plain-language recap, a jargon decoder, and a pre-visit Prep Sheet for next time.
Cross-source reconciliation
The differentiator. DiagNova does not just store three data types side by side — it reconciles them. Claude cross-checks the spoken care plan against wearable trends and uploaded labs, flagging tensions for the patient to raise with their doctor, rather than deciding anything itself.
Where a wrong answer is most dangerous
The reasoning across these silos is where the product lives or dies. Core Defender evaluated several frontier models and standardized on Claude on Amazon Bedrock for three reasons.
Honesty under uncertainty
Whether reading a smudged lab PDF or an ambiguous transcript, Claude was markedly more willing to say “this value is unclear, confirm with your provider” than to fabricate a confident answer. For a patient-held record, a calibrated “I’m not sure” is a feature, not a failure.
It holds a safety line
Claude reliably stayed inside DiagNova’s “inform, never diagnose” boundary across every surface — explaining what a result means while declining to tell a user what they have.
HIPAA-eligible, no training on customer data
Running Claude through Amazon Bedrock let DiagNova meet its HIPAA obligations contractually, with PHI never used to train models — a non-negotiable for shipping in healthcare.
“We're a small team handling people's most sensitive data. We needed a model that knows the edge of its own knowledge and refuses to guess. That is the difference between a tool a doctor tolerates and one they trust their patients to use.”
What it added up to
Patient impact
A visit recap delivered within 90 seconds of the appointment ending.
Surfaced per visit by the Prep Sheet, drawn from the patient's own wearable and lab trends.
To turn an uploaded lab PDF into structured, trended biomarkers — versus manual entry.
Business impact
By building on Claude instead of training and certifying a medical reasoning model — DiagNova shipped its flagship reconciliation feature in one quarter.
Per active user, after model tiering — routing high-volume extraction to smaller Claude models and reserving Opus for visit reconciliation, with no measurable quality drop on the hard tasks.
Building something in healthcare?
We help teams ship safe, HIPAA-eligible AI on Claude and AWS Bedrock — from document ingestion to clinical reasoning that knows the edge of its own knowledge. Let's talk.
Contact Us