For hospital insurance desks in India

Cashless pre-authorisation, drafted in minutes instead of hours.

Upload a patient's clinical file, pick the insurer, and get a TPA-ready pre-authorisation letter with a gap analysis and confidence-scored extracted data — then have a person check it before it goes anywhere.

Create an account Sign in

New accounts wait for a hospital administrator to grant access. Patient data is never visible until someone assigns it.

Why a claim gets queried

A cashless pre-authorisation is not one form. It is a request, a query loop, an enhancement when the bill grows, a final authorisation at discharge, and a settlement months later — and most of it is assembled by hand, under time pressure, from a paper file.

The request is built by hand

A coordinator reads the case notes and retypes the clinical picture into the insurer's form. The same facts, transcribed again, at the end of a shift.

Queries cost days

A TPA query on a missing document or a thin justification restarts the clock. On an admitted patient, that delay is a bed and a discharge held up.

Nobody can reconstruct it

When a claim is disputed weeks later, the question is what the hospital actually told the insurer and when. Usually the answer is in somebody's memory.

How it works

Four steps. A person is required at the last one, by design.

Step 1
Read the file
Upload the discharge summary, investigations and prescriptions as PDFs or photos. Text is extracted in the browser — digital PDFs directly, scans and photos through OCR.
Nothing is sent to a third-party document processor.
Step 2
Extract, with a score
Diagnosis, procedures, medications and ICD-10 codes come back as structured fields — each with its own confidence score, not one number for the whole document.
Below 60 is flagged and must be corrected.
Step 3
Draft the letter
A pre-authorisation letter addressed to the insurer, with the medical-necessity justification grounded in the specific findings in the file, plus a gap analysis against that insurer's own document list.
Prints on the hospital's letterhead, ready to sign.
Step 4
A person approves it
Nothing is transmitted. The desk reviews every flagged value, the treating doctor signs, and a person records that it went out through the hospital's own channel.
The software never contacts an insurer.

How it holds up when someone checks

An AI tool that drafts a clinical document for an insurer is only worth deploying if it can be audited afterwards. These are mechanisms, not intentions — each one is enforced in the database, where the application cannot talk its way past it.

The gates are in the database, not the interface

A claim with an unverified value cannot be marked ready to submit — a Postgres trigger refuses the write. Disabling a button in the browser would be a suggestion; this is a rule that holds even for a request made straight to the API.

“Not checked” never looks like “checked and clean”

When an insurer's document requirements are not on file, the claim says documentation was NOT checked, in those words. An empty gap list rendered as a green tick is how a hospital learns to trust an analysis that never ran.

A human correction is recorded as a human correction

The audit trail distinguishes what the model extracted from what a person fixed, and says so plainly when it cannot tell. The one event that proves a human checked the AI is worth recording accurately.

Every state change is append-only

Who did what, when, and from what state to what state — in tables no user can update or delete, including administrators. That ledger is what a hospital shows an insurer when a claim is disputed.

Built for Indian rules, not adapted from American ones

DPDP Act 2023

India's data protection law governs this product, not HIPAA. There is no “BAA” in Indian law; the equivalent is a Section 8(2) processor contract. Health data is personal data and is handled uniformly.

Data stays in India

Patient records live in a Mumbai region database. Moving the project out of India would require a Section 16 cross-border review, and the deployment is pinned accordingly.

TPA vocabulary

Insurer document lists are per-hospital, because empanelment differs by hospital and TPA desk. A shared default is never silently applied as if it were yours.

Where this is today

PreAuth is pre-launch and runs on synthetic patient data only. The intake, extraction, letter generation, query resolution, audit trail and administration are built and working. Eligibility checks, enhancement requests and settlement tracking are not built yet — they are the next part of the claim lifecycle, and they are named here rather than implied.

A list of the data-protection obligations still outstanding is maintained in the repository and reviewed before each phase. None of it is optional before production.

See it with your own file

Create an account and a hospital administrator can grant you access. Use synthetic records — this is pre-launch software and should not hold a real patient's file yet.