A compliance-grade multi-agent platform whose every decision can be reconstructed — and verified by an auditor who trusts nothing but the mathematics.

The challenge

When a regulator audits a claim months later, how do you prove it was justified at the time it was billed — with evidence an adversarial third party will accept without having to trust the system that generated it?

What we built

Four specialised agents run in sequence under orchestration — consent compliance, claims analysis, forensic review, attestation authoring — each emitting a hash-chained receipt the next agent verifies before proceeding. A redaction boundary after the first agent means the model never handles patient-identifying data downstream. The pharmacist stays the legal trust anchor: the pipeline pauses at a briefing card and nothing seals without an explicit human decision. The cryptographic path contains no LLM at all — HSM-backed ECDSA signing, trusted timestamps, an append-only SQL ledger — and an anonymous in-browser verifier lets an auditor re-check any packet against a published public key, with no backend and no login.