Watch an AI agent review an insurance claim, step by step

A teaching demo of agentic AI: upload a claim, watch the agent extract the data, retrieve the relevant rules, validate them one by one, and explain its decision — with every step of its reasoning visible.

All data is fully synthetic. No real patient information is used anywhere.

  1. 1

    Extract. The model reads an unstructured claim document and pulls out structured fields.

  2. 2

    Retrieve. It searches the policy rules for the ones most relevant to the claim (RAG).

  3. 3

    Validate. It checks the claim against each rule — pass, fail, or unclear — with reasoning.

  4. 4

    Decide. It synthesizes a verdict (approve / deny / needs review) and explains why.

  5. 5

    Evaluate. Score the whole labeled set against ground truth with a confusion matrix.

Review claims →EvaluationEdit prompts