YBA Med — Layer 2 Specialist Agent Design

What this is: the full Layer 2 specialist agent fleet for YBA Med, designed by 5 parallel research specialists on 2026-05-02. Sits ON TOP of the 8 universal Layer 1 foundation agents, with the Layer 3 Richards-Brain markdown vault per tenant.

Total: 42 specialist agents across 5 teams, each agent with defined role / skills / SOPs / Layer 1 interactions / hard NEVER rules.

The 5 Teams

Plus the 8 universal Layer 1 agents (Infrastructure, Development, HR, Accounts, Front Desk, Customer Care, Call Transcription, Executive Assistant) → 50 agents total per Med tenant.

The full per-team specialist briefs from each specialist are preserved below.

Team 1 — Clinical Workflow (9 agents)

These are the agents that ride with the doctor through every consultation.

The hard rule across this team: agents draft, doctors sign. Every diagnosis, prescription, referral is doctor-approved before action. NAPPI/ICD-10/drug-interactions come from verified lookups, NEVER from LLM generation.

Team 2 — Billing & Medical-Aid Claims (7 agents)

This is the moat. Claims processing is what doctors complain about most.

The Rejection Auto-Fix algorithm (the moat): 1. Classify rejection into ~30 known categories (NAPPI/ICD mismatch, missing field, exhausted savings, prior auth needed, etc.) 2. Apply playbook per category 3. Re-submit if fixable; escalate to human if not 4. Three failed auto-fixes = mandatory human escalation 5. Trust ratchet: 200 successful auto-submissions of a claim type → that type unlocks for auto-submit-without-approval

What this saves a 2-doctor GP practice: R50-85k/month in recovered revenue + freed staff time. Justifies R20k/mo subscription many times over.

Team 3 — Patient Lifecycle & WhatsApp (10 agents)

Patient journey from first Google search to long-term retention. WhatsApp-first because that’s what SA patients use.

Critical sub-systems: - WhatsApp Portal uses 360dialog (or Infobip) — the WhatsApp Business API, not the free app. Tracks the 24h customer-care window, language detection, opt-out enforcement. - Red-flag triage is uninterruptible. Trigger phrases (clinically reviewed): chest pain · can’t breathe · bleeding heavily · stroke signs · paediatric red flags · suicidal ideation · trauma. Multi-language. Single safe reply: “This sounds urgent. Call 10177 or 112 NOW. Alerting the doctor — please stay on this chat.” Telegram + SMS + voice call to on-duty doctor in 30 seconds. Weekly synthetic test messages.

What this changes: No-show rate drops from 15-25% → 3-6% (R40-80k/month recovered). Front-desk phone load drops 60-70%. Patient satisfaction measurable for the first time.

Team 4 — Compliance, Records & Safety (9 agents)

Layer 1 HR has BCEA/LRA. Layer 1 Accounts has SARS/VAT. Med Compliance team adds the medical-specific regulatory layer.

The compliance brain structure (markdown vault):

wiki/compliance/
├── hpcsa/practitioners/<doctor>.md   # registration, scope, CPD ledger, indemnity
├── bhf/practice-pcn.md
├── popia/{information-officer,breaches,subject-access-requests}/
├── controlled-substances/{register-current,monthly-counts}/
├── infection-control/{autoclave,sharps,incidents}/
├── insurance/{professional-indemnity,public-liability}/
└── calendar.md                       # the master annual compliance calendar

Hard bright lines (no exceptions): Agents NEVER auto-submit to HPCSA, IR, SAHPRA. Patient records never deleted before retention expires. Audit log is append-only — even the agents can’t edit it. The doctor signs every release of a record, every contract, every complaint response.

Team 5 — Practice Operations (7 agents)

Layer 1 HR handles employment law in general. Med Operations team adds medical-practice-specific resource management.

Disaster scenarios + agent response:

What this saves the owner: 15-20 hrs/week of admin (~R28-35k/mo of practice-manager time freed); R30-80k/year vaccine wastage avoided; R15-25k/year stock-out savings; zero missed equipment services.

Coverage Map — every aspect of running a SA medical practice

Hard Safety Rails — the universal NEVER list

Across all 5 Med teams:

1. No autonomous clinical action. Every diagnosis, prescription, referral is doctor-signed before action.
2. NAPPI / ICD-10 / drug interactions from verified lookups only — never LLM-generated.
3. Red-flag triage is uninterruptible. Cannot be disabled by code, prompt, or instruction.
4. Patient never sees a result before the doctor reviews it.
5. POPIA special-category rigour. Health data tighter than ordinary PII. Per-record audit trail.
6. HPCSA records retention auto-enforced. 6 yrs adult, until 21 minor. Cannot hard-delete before then.
7. Audit log is append-only. Even agents can’t edit it.
8. Multi-tenant isolation. Practice A’s brain never leaks to Practice B. Niki audits every cross-tenant boundary.
9. No agent ever auto-submits to a regulator (HPCSA, IR, SAHPRA, BHF) — doctor signs.
10. Practice owner has a single kill-switch. One Telegram command pauses all outbound + escalates to humans only. Tested monthly.

What this means for the empire architecture

1. Layer 2 for Med = 42 specialist agents in 5 teams (in addition to 8 Layer 1 universal agents).
2. Other verticals (Edge, Terra, Realty) will have their OWN Layer 2 fleets with totally different agents, but the same 5-team shape (clinical/operational/customer-facing/compliance/operations) probably won’t apply uniformly. Each vertical’s specialists fit that vertical’s reality.
3. Foundation Layer 1 doesn’t change between verticals — the universal 8 are universal.
4. Layer 3 (markdown brain) is the same shape — Richards Brain pattern per tenant, with vertical-specific subdirectories under wiki/.
5. For Concierge tier pricing: Med tenants should be priced at the upper end of the R18-30k/mo range (call it R25-35k/mo for a multi-doctor practice) given the regulatory-load + revenue-recovery upside. The Rejection Auto-Fix Agent alone justifies the price.

— Synthesised by Niki on 2026-05-02 from 5 parallel research specialist outputs.