Reimbursement claims, reconciled — not just extracted
The claims desk that knows when to stop.
Munshi reads a patient's handwritten Hindi bill, discharge summary and policy with two Sarvam models, certifies only what they agree on, and prepares a submittable bilingual packet — with the math in code and a refusal where it can't be sure.
For patients
Snap three photos, from your phone
Open your hospital's link, photograph your bill, discharge summary and policy — or just type the policy number — and track your claim in Hindi or English. No login, under three minutes.
Open a demo intake link →For the hospital desk
One worklist, both entry points
Patient submissions and walk-ins land in the same queue. Triage incoming claims, resolve the one contested field, watch the payable cascade with cited deductions, and hand off a bilingual packet.
Enter the console →How Munshi works
Certify what's true. Refuse what isn't.
01
Read every document
Sarvam Doc AI digitises the handwritten Hindi bill, discharge summary and policy — text plus block coordinates.
02
Two models, independently
105B reads as primary; 30B challenges. Munshi certifies only what they agree on and refuses — CERTIFIED, CONTESTED, or UNREADABLE — what it can't verify.
03
The arithmetic, in code
Deterministic math computes the expected payable with every deduction cited to a policy clause and a bill line. The model never touches a number.
04
A submittable packet
A bilingual packet an operator can hand to the insurer — with a Hindi read-aloud and a full audit trail. It prepares; it never files.