Health records · AI-assisted · non-diagnostic
HealthTrack turns raw lab data into something you can actually read — with trends over time and grounded guidance that assists, but never diagnoses.
Your HbA1c has risen steadily over the past year and is now above the normal range. It reflects average blood sugar over roughly three months, and is worth discussing with your doctor.
This is context, not a diagnosis. The trend and reference range are computed deterministically; the wording is drafted by an AI and reviewed before it reaches you.
Access control
HealthTrack isn’t a single dashboard. It’s a permissioned system where every role has its own view and its own boundaries — enforced on the server, deny-by-default.
Owns their record. Sees results, trends, plain-language guidance, and reminders — and controls exactly what’s shared.
Sees only results a patient has explicitly shared. No trends, no advisor — access is scoped and revocable.
Uploads results, drafts summaries with AI, and extracts measurements from real lab PDFs with human review.
Operational only — approves labs, verifies licenses, views anonymized analytics. Never holds clinical data.
AI that assists
Explains what a result means and drafts follow-up reminders. Trends and flags are computed deterministically — the AI only writes the explanation, so it can’t invent a clinical value.
Reads measurements from real lab reports (text layer or OCR), structures them against known analytes, and shows the lab exactly what it read — to correct before confirming.
Whole-picture guidance is drawn from your own results over a window you choose — never generic advice — and always framed as non-diagnostic.
Built with care
The system explains and contextualizes; it never diagnoses. Deterministic trend analysis is kept separate from AI-written text to eliminate hallucinated clinical values.
Flags are computed against independently-sourced clinical ranges — a lab can’t define its own “normal” and quietly flag away an abnormal result.
Roles are strictly bounded and enforced server-side. An admin operates the system, but can never be assigned clinical results.
This is a portfolio build running entirely on synthetic records. No real patient data is involved.
Grounded in real sources
The advisor doesn't invent medical explanations. Every “what this means” is grounded in a curated knowledge base — 30 sourced passages from MedlinePlus, WHO, CDC, Mayo Clinic, Cleveland Clinic, and NORD, covering diabetes, cholesterol, hypertension, anemia, sickle cell disease, malaria, and typhoid.
Five core analytes — HbA1c, Total Cholesterol, LDL, HDL, and Triglycerides — are compared against reference ranges read directly from MedlinePlus and Johns Hopkins Medicine, one citation per number. Other numeric results are extracted and shown, but flagged “unknown” rather than guessed.
Rapid, titer, and graded tests (malaria, typhoid, and others) are flagged against the assay's own reported cutoff or expected value — never a lab-invented range. New sourced ranges are added over time; accuracy comes before coverage.
Create an account and explore the patient experience, or sign in.