Find the right evidence.
Semantic search and keyword retrieval work together, followed by reranking, to surface the policy clauses that matter to the case.
Behind every decision.
A reason you can trace.
Health-insurance claim analysis, grounded in the policy.
Five specialist agents. Cited evidence. Reasoning you can inspect.
THE SOURCE OF TRUTHThe policy.
The policy.
At the center.
A STRUCTURED RESULT
Supporting policy citations
Evidence confidence breakdown
Validation & review status
Follow the evidence from the policy to the finding.
The conclusion matters.
So does how you got there.
The engine uses the indexed policy as its source of truth. It retrieves relevant clauses, examines coverage and exclusions, and checks whether the findings are actually supported.
FIVE SPECIALISTS. ONE CONNECTED WORKFLOW.
From case to clarity.
HOW IT WORKS / 01
First, understand the case.
Extract the claim facts and identify what needs investigating. The case analysis agent frames questions without drawing policy conclusions.
Interactive workflow guide · Open the live app to analyze a case
Inspect the reasoning.
Explore the agent trace, citation inspector, and confidence breakdown in the live interface.
Know when to review.
The validation stage checks claim-to-evidence support. Missing or unresolved evidence can return a needs-review result.
See the evidence.
Understand the decision.
Start with a preloaded test case, upload a JSON case, or paste one into the interface. Then follow the analysis.
Try Claim Engine Explore the sourceA CLOSER LOOKGood questions.
Good questions.
Clear answers.
What can I try in the demo?
Select a preloaded case, upload a JSON case, or paste one manually. The app displays the analysis, agent trace, supporting citations, and confidence breakdown.
Which policy does it analyze against?
The repository indexes policy UNIHLIP18004V011718. Results depend on that policy and the case facts supplied; this is not a tool for comparing every insurance policy.
What does NEEDS_REVIEW mean?
The engine cannot support a reliable conclusion from the available evidence. It flags the case for review instead of forcing a coverage decision.
Is this a final insurance determination?
This is a technical demonstration of policy-grounded analysis. Its output does not replace an insurer’s claim determination or professional review.
Why might the Hugging Face app take time to open?
A hosted Space may need to wake up or load its models. The Hugging Face page shows its current runtime status.
Built by Aniket Chavan · Python / LangGraph / FastAPI / StreamlitExplore all apps