BillRadar: An AI Agent That Checks a Whole Pile of Hospital Bills
Built in-house
Retailbox R&D: built in-house for the Nebius × NVIDIA Global AI Hackathon, 2026
An agent that reads the hospital's bills, the doctors' bills and the insurer's statements together, finds billing errors such as duplicate charges, checks the hospital's own financial assistance policy and drafts the letters. Only the family sends anything.
What was the problem?
After a hospital stay a family gets a pile of paper: bills from the hospital and the doctors, the lab, letters from the insurer. Errors hide between the pages. The same anesthesia is billed twice. A newborn's care is billed under the mother's insurance ID. And the financial assistance application has a deadline.
How did Retailbox solve it?
The family photographs the table with all the papers, plus any PDFs. BillRadar finds and straightens every sheet. On the device, it covers names, member and account numbers, addresses, claim and statement numbers and barcodes before anything goes to a model, and checks each covered sheet again; a sheet that still shows one of these is not sent. It reads every line as printed, links pages from different letters into one case, and checks the case against the hospital's own financial assistance policy and federal rules, quoting each rule with its page. If the hospital isn't in its base, it finds the policy with Tavily, searching only the hospital's own site. Code recomputes every amount and date and holds back anything it can't confirm. Then Nemotron 3 Ultra reads the whole case again and flags what doesn't add up: a flagged finding moves to 'worth checking' without a dollar amount. It drafts the letters to sign and fills in the hospital's real assistance form.
The models read and explain, and plain code decides whether there is money. Each step gets the model size it needs, chosen by measuring on fictional bill packs. Other AI agents can use BillRadar over MCP and A2A.
Every bill in the demo is fictional, and no real bills have been sent to any model. The example family and hospital are made up; the financial assistance rules and form are the real public ones of a Florida hospital, applied to the fictional case. Versions of BillRadar were also entered in the Meta Global AI Developer Hackathon and in Flip the Switch.
Facing the same problem? These are the services that cover this kind of work: AI agent development services.
What does it look like?
(opens the full-size image in a new tab)
(opens the full-size image in a new tab)What were the results?
- ✓A whole example case, 14 sheets in 3 files, read and checked with 15 Nemotron calls for about 15 cents ($0.146; translating the letters is extra)
- ✓The second read by Nemotron 3 Ultra doubted 6 of 8 wrong findings, where Nemotron 3 Super doubted none; it also doubted 4 of 9 right ones, which stay on the page under 'worth checking'
- ✓Letters translated by Nemotron 3.5 Lightning: 53 of 56 parts correct against 54 of 56 on Super, at about 5.5 times lower cost
- ✓Policy search tested on 5 real hospitals in 4 states: a policy was found for all five, one was confirmed in full with a revision date, and anything unconfirmed is shown as a link to check