Nivaran Claims Desk
Catch the deduction before the claim leaves the hospital.
Nivaran reads the whole insurance claim file, checks it against the patient's policy, calculates what the insurer will actually pay, and drafts the paperwork. AI reads the documents; code does the math.
Every TPA desk fights the same battle
Numbers from 18 months at the fictional Sahyog Multispeciality Hospitals, computed live from this demo's data.
₹2.99 Cr
lost to room-rent rules
A room above the policy cap shrinks nursing, surgeon and OT fees in proportion. Families find out at discharge.
24%
of files get a TPA query
A missing first-consultation note or implant sticker sends the file back and delays payment.
15 days
median wait to settle
Every query adds days. Getting the file right the first time is the fastest fix.
How it works: one file in, nine visible steps, one answer out
AI reads, classifies and writes. Calculated steps are rules and arithmetic, tested to the paisa, with the formula shown.
Where the AI works, and where it does not
AI
Reads scanned pages
Sorts 14 mixed pages into documents and reads names, dates and every bill line, including photocopies.
See it in the tourAI
Shows its source
Every value keeps the page it came from; click it and the viewer highlights the line.
See it in the tourAI
Takes a second look
Before flagging a mismatch on a scanned page, it re-reads that value at higher detail.
See it in the tourAI
Codes the diagnosis
Suggests the ICD-10 code from an allowed list, with its confidence and alternatives.
See it in the tourCalculated
Applies the policy
Waiting periods, room caps, sub-limits and co-pay, from versioned rule files.
See it in the tourCalculated
Calculates the payable
Proportionate deduction and every other rule, to the paisa, with the formula visible.
See it in the tourCalculated
Ranks the fixes
What to correct before submission, ordered by rupees saved.
See it in the tourAI
Drafts the paperwork
Cover letter, TPA query reply citing pages, and a note for the family in Gujarati or Hindi.
See it in the tourCalculated
Degrades gracefully
No API key or over budget? The same screens run on stored answers or rules.
See it in the tour
The demo runs on a full hospital's worth of data
- Claims
- 11,050
- Bill lines
- 4,50,000
- Admissions
- 11,050
- Insurer settlements
- 9,900
Fictional and deterministic. It resets every night at 3:00 AM IST so the demo is always current.
Built with
- Python 3.13
- FastAPI
- Celery
- PostgreSQL + pgvector
- Redis
- HTMX
- Alpine.js
- Tailwind CSS
- ReportLab
- PDFium
- OpenAI GPT on Microsoft Foundry
- Azure Container Apps
A FastAPI app renders the desk with HTMX; a Celery worker runs each claim's nine steps and streams progress to the browser. Every model call goes through one gateway with live, cached and rule-based modes, a daily budget and a full trace. The money engine is plain Python on Decimal, driven by versioned rule files and unit-tested to the paisa.
Built by AIGyde Solutions
We build AI products that read the messy documents and keep the important numbers in tested code. If your team spends its day checking files against rules, we should talk.