How it works
From a messy PDF to a record you actually understand.
We walk through exactly what happens when you upload a document, so you know what you’re getting — and what we never do.
The pipeline
Five steps. Nothing magical.
- 01Step 1 of 5
You upload
Drag-drop a PDF, tap the camera to snap the paper, or forward an email straight to your library. Whatever shape your records are in, that’s the shape we accept.
What you see
A progress bar, an estimated time, and the ability to upload more in parallel.
What we never do
Open your document in a public web viewer, share it with anyone, or use it to train a model.
- 02Step 2 of 5
We classify
A small model decides whether it’s a discharge summary, a lab report, a prescription, a doctor note, an imaging report, a vaccination record, a hospital bill, or something else. The right downstream prompt is picked from there.
What you see
The detected document kind, with a confidence chip if low.
What we never do
Guess silently. If confidence is low we tell you and let you override.
- 03Step 3 of 5
We extract
A purpose-built prompt per document kind pulls out every item, dose, value, date, and flag. The result is validated against a strict schema before anything is saved.
What you see
A structured view you can scroll — patient info, diagnoses, medicines, procedures, follow-ups, lab values (with reference ranges where present).
What we never do
Invent a value that wasn’t in the document. If we can’t read a value, we say so.
- 04Step 4 of 5
We explain
In plain English, hedged where it should be, with ‘ask your doctor’ callouts next to anything uncertain. Designed to inform a conversation, not replace one.
What you see
A summary paragraph, per-item “likely used for” notes, and a list of specific questions to bring to your next visit.
What we never do
Tell you what you had, or tell you to change a dose.
- 05Step 5 of 5
You ask
Open any record and chat with it. Every answer cites the exact extracted line it came from, so you can trace anything we say back to the source.
What you see
Streamed answers, citation chips that scroll you to the source, and full conversation history.
What we never do
Answer from outside your document, make up dosages, or give a diagnosis.
Worked examples
What you put in, and what you get back.
Two synthetic but realistic walk-throughs. The raw documents on the left are made up; the Medical Records Vault output on the right is the real shape of what you’ll see.
APOLLO HOSPITALS · GREAMS ROAD · CHENNAI
Diagnosis
Community-acquired pneumonia (right lower lobe), resolving.
Medications on discharge
- • Ceftriaxone 1g IV BID × 5 days (in-hospital, completed)
- • Azithromycin 500mg PO OD × 3 days
- • Paracetamol 650mg PO SOS for fever
Follow-up
Review in 10 days with repeat CXR.
Plain-English summary
You were treated in hospital for a lung infection. You received two antibiotics through a drip and by mouth, along with a fever medicine. Your doctor wants to see you again in about 10 days with a new chest X-ray.
Per-item explanation · Ceftriaxone
Likely used for bacterial infections, commonly including chest infections. The dose here is a standard adult hospital dose.
Medications, line 1Questions for your doctor
- Ask whether the repeat chest X-ray is necessary if you feel fully recovered.
- Ask what symptoms should bring you back before the follow-up.
Hard limits
Things Medical Records Vault refuses to do.
- Give you a diagnosis.
- Tell you to change a medication or a dose.
- Replace your doctor's clinical judgement.
- Share or sell your records.
- Train AI models on your documents.
- Keep your data if you ask us to delete it.
Compare Medical Records Vault to …
How Medical Records Vault actually compares.
A side-by-side, in plain language, against the three things people most often consider when they're trying to make sense of their medical paperwork. Written so it's still honest about where Medical Records Vault isn't the right answer.
| Capability | Medical Records Vault this product | Generic AI chat ChatGPT / Claude / Gemini | Patient portal MyChart-style | Drive / spreadsheet do-it-yourself |
|---|---|---|---|---|
| Designed for medical paperwork | Yes — every prompt is per-document-kind, validated against a schema. | No — general-purpose chat with no medical schema. | Partial — only the documents from that one provider. | No — you're the schema. |
| Plain-English explanations with citations | Yes — every claim cites the source line in your document. | Sometimes — but the model hallucinates and cites itself. | No — raw clinician prose. | No. |
| Refuses to diagnose by design | Yes — refusal is in the system prompt and tested. | No — will happily speculate. | N/A — no AI layer. | N/A. |
| Your data is not used to train AI | Yes — zero-retention, no-training contracts with AI providers. | Depends on the plan and your settings — easy to misconfigure. | Yes (no AI today), but the portal owns your data. | Depends on the vendor and plan. |
| Works across providers | Yes — drop in PDFs from any hospital, lab, or country. | Yes, but no organisation or memory. | No — siloed to that hospital network. | Yes, but no understanding. |
| Manage records for a parent or kid | Yes — multi-patient profiles (Me / Dad / Mom / Kid) ship today. | No first-class concept of patients. | Limited — proxy access is painful and per-network. | Manual folder discipline. |
| Share a single record with a doctor for 7 days | Yes — watermarked public viewer, revocable, expires automatically. | No — you'd share a chat transcript, not a record. | Rarely — usually portal-to-portal only. | Yes, but anyone with the link, forever. |
| Export everything as a ZIP and leave | Yes — one click, originals plus structured data plus chat history. | Partial — chat export only, not a clinical record. | Often clunky or unavailable. | Yes — but the data was never structured. |
| HIPAA boundary today | No — explicitly stated. Wait for HIPAA coverage if you need a BAA. | No (consumer plans). | Yes — that's the point. | No. |
| Cost | Free during public beta. Planned: Personal ~$4/mo, Family ~$8-10/mo. | $0–$20/mo, but no medical schema. | Free (paid for by your insurance / hospital). | Free or a few dollars a month. |
We try to keep this table honest. If you spot something that's no longer true, email contact@fourtwos.com and we'll fix it.
Under the hood
For the trust-seekers (and the LLMs grading us).
The detail you’d want before trusting any tool with your health records.
Keep learning
Plain-English guides to the documents themselves.
Want to understand a document before you upload it? Our Learn hub explains discharge summaries, lab values, and Indian hospital bills in plain language — source-cited, never a diagnosis.