MiddyMindEmpowerReg
Fuses MAUDE reports, recalls, complaints, service logs, and EHR adverse events into one device Safety Graph - so hospital procurement and quality teams see harm signals before they cascade.
Every signal is tied to a resolved device entity with per-source freshness and provenance. The agent decides tone; MiddyMind decides what it can see.
Step-by-step
4 steps- Step 01
Ingest VOC + MAUDE
You do → Upload the customer_vocs sample and let the MDR/QMSR completeness gate score each row.
You'll see → A gated intake tracker: rows with missing required fields flagged red; passable rows advance to the resolver.
Under the hood → identity/ resolves manufacturer + device model + lot across sources.
- Step 02
Open Safety Radar
You do → Pick a device family (infusion pump) and expand its safety card.
You'll see → MAUDE events, service tickets, and internal complaints collapsed onto one device with a bitemporal timeline - 'what was true, when'.
Under the hood → memory/ answers the temporal query; every fact carries a source id.
- Step 03
Read the Signal Brief
You do → Hover a case to see the ~200-token brief the triage agent will see.
You'll see → Tier, open items, promises, and top facts with per-source freshness - plus the fact IDs the agent must cite.
Under the hood → conductor/signal_brief.py compiled fresh from memory this turn.
- Step 04
Escalate with autonomy tier
You do → Set the Patient Safety Radar agent to one-click-approve and run it against yesterday's batch.
You'll see → Draft escalations queued for approval, each with a cited rationale and the confidence label.
Under the hood → governance/ enforces the tier, cost cap, and PII redaction; append-only audit trail.
Shipped in this vertical
- customer_vocs schema + intake trackers (migration 0015)
- 3 seeded agents: Safety Radar · VOC ingestion · Patient Safety Radar
- MDR/QMSR completeness gate