@devansh_vashishat_45
1
Events attended
1
Submissions made
Looking for experience!
Socials
š¤ Top Collaborators
š¤ Latest Submissions
Juro - Both sides heard. One ruling.
In the US, about one in five in-network health claims gets denied, and fewer than one in a hundred is ever appealed. The denials usually do not win because the law is against the patient. They win because nobody builds the case. Juro builds it. When a claim is denied, four AI agents take it to a small tribunal in one shared room on Band. An Advocate argues for the patient and pulls the records and policy language that support coverage. A Scrutinizer argues the insurer's side and tries to poke holes in the claim. Evidence stays neutral and produces the exact document or statute that settles a point. An Adjudicator runs the room and recommends a ruling. Then it stops. A human reads the whole debate and delivers the binding decision, and Juro drafts the appeal letter with the citations already in it: ERISA's right to a full and fair review (29 CFR 2560.503-1) and the ACA's external review (45 CFR 147.136). The agents do not follow a script. Each is a separate process that connects to Band over a WebSocket and argues by answering the others in that one room. The panel runs on two models on purpose: the debaters on a fast one, the chair on a heavier one, because writing the ruling is the hardest call. Getting four models to hold a clean debate was the hard part, and two things kept breaking it. A model would mention the wrong agent and the hearing would stall, so we force every handoff in code. And an agent would reason out its ruling but forget to post it, so we capture the text and post it anyway. Every turn is hash-chained into one root, so the record cannot be quietly edited. Health insurance is the wedge because the rules are already written down. The same engine fits disability, veterans' benefits, and prior authorization next.
19 Jun 2026
