Back to the discharge queue
Kiran Patil, claim C-106
What this demo claim shows: An Rs 9,500 MRI has no order or report, and a lab report matches another patient’s claim. The agent flags both and does not try to fix them.
Not run yet
1. Claim file
Synthetic patient and documentsInsurer and policy
Sahyog General Insurance (fictional), Arogya Shield Plus, sum insured Rs 5,00,000
Policy since
2025-12-01
Room
General ward, Rs 3,000 a day for 3 days
Desk note
Bill finalised late last night
Documents in the file
2. Bill lines and evidence
Not checked yet| Line | Description and code | Amount | Evidence |
|---|---|---|---|
| L1 | General ward, 3 days ROOM-GW | Rs 9,000 | Not checked |
| L2 | Physician visits x3 PRF-VISIT | Rs 2,400 | Not checked |
| L3 | Fever panel: CBC, malaria, dengue NS1 LAB-FEVER-PNL | Rs 2,100 | Not checked |
| L4 | MRI brain (plain) RAD-MRI-BRAIN | Rs 9,500 | Not checked |
| L5 | Pharmacy (itemised) PHA-IP-412 | Rs 5,200 | Not checked |
3. Policy checks
Deterministic rules engine, not the language modelThe rules run when the agent calls the policy tool. Each result cites the clause it comes from.
4. Agent trace
0 tool callsLLM planner, gpt-4.1-miniThis deployment has a model configured. The language model will choose each tool call; you will see its reasoning notes.
Press Run the agent. Every card is a real tool call with its input and output; nothing is pre-recorded.
7. Messages with the insurer
Hospital and insurer go back and forth here until the claim is approved or rejectedLoading the conversation.