Bring your own model
Pick Claude, GPT, Gemini, or Llama per workspace. Switch anytime; settings are versioned.
Feature · AI architecture
Useful AI. Auditable AI. Your AI.
MedIQ is AI-native — but we never bet patient outcomes on a single model. A deterministic rule engine does the heavy lifting; your chosen LLM adds judgment on top.
Bring your own key
Your OpenRouter key is encrypted AES-256-GCM at rest with per-record IVs and tenant-scoped access.
De-identified prompts
We strip patient identifiers before the model call. Prompts log only an input hash.
Auditable decisions
Every AI call writes a row: model, version, input hash, decision, recommended fix.
Deterministic-first
The rule engine runs before the model. If a rule blocks, we never ship the prompt.
Latency budget
Scrubber AI pass typically completes in under 1.5s p95 — fast enough for inline review.
Pipeline
A claim's trip through the AI layer.
Plain-English description of the path every claim takes — deterministic checks, then the LLM, then a recommendation your biller approves.
- 1IngestClaim arrives from encounter or batch upload. Tenant + role checked.
- 2Rule engine60+ deterministic rules: crosswalks, modifiers, duplicates, payer rules.
- 3De-identifyPHI stripped; structured payload + input hash prepared.
- 4Model callYour chosen model returns a risk score + reason.
- 5RecommendWe synthesize rule + AI output into a fix the biller can accept in one click.
- 6AuditDecision + model + input hash written to the audit log.
Want the technical brief?
We'll send the architecture overview, evaluation harness, and current model benchmarks.