Plan — AZ One Health Sentinel

A working name for the application this plan describes: a mobile-first participatory-surveillance + early-warning app for Arizona that operates across two flagship verticals — Vector-Borne Disease (VBD) and Heat — backed by the existing DuckLake knowledge graph and driven by an agentic architecture that consumes real-time data via MCP servers.

What you’re reading

  1. 01-parameter-mapping.md — the Figure 2 Minimum Set of Key Data Parameters mapped to each vertical: which parameters apply, which are additions, which are suppressed, and how each parameter lands in DuckLake.
  2. 02-mcp-integration.md — the MCP server inventory (one built, eight to build) with a parameter-by- parameter table showing which API feeds which Minimum-Dataset slot.
  3. 03-agentic-architecture.md — the eight-agent topology that turns a community report into a triaged, enriched, validated observation in the knowledge graph and a personalized response back to the user.
  4. 04-data-flows.md — four worked end-to-end scenarios (tick mailed in, sick mosquito-bitten patient, unsheltered heat-distress check-in, agency-side cluster detection).
  5. 05-roadmap.md — phased delivery: hackathon MVP → 6-month pilot → 12-month evaluation against the Figure 3 timeliness milestones.
  6. 06-mobile-app.md — anonymous-first Human / Animal / Environmental reporting app: low-text UI, EXIF-stripped photos, ZIP/km coarsening, “never diagnose” risk boundary, DuckLake snapshots backed by GitHub LFS, web pilot → iOS + Android.
  7. 07-auth.md — optional account system (email/password, magic-link, OAuth via Google / Facebook / Apple) layered on top of anonymous reporting. Supabase Auth backend; right to erasure detaches but does not delete attached observations.

The thesis in one paragraph

The Minimum Dataset (Figure 2) is the shared data contract that lets one app serve both verticals: the General class is identical for VBD and Heat, the Human class diverges into infectious symptoms vs. heat-illness symptoms, the Exposure and Environmental classes pick up domain-specific sub-fields, and the Livestock / Wildlife classes matter for VBD but are silent for Heat. Treating the dataset as a contract — rather than building two apps — means one knowledge graph, one set of MCP integrations, one agentic pipeline, and two thin domain-specific UIs on top.

The architecture in one diagram

   ┌──────────────────────────────────────────────────────────────────┐
   │                       Community user (mobile)                   │
   │      VBD UI · Heat UI · SMS fallback · Voice (transcribed)      │
   └─────────────────────────────────┬───────────────────────────────┘
                                     │  free-text / photo / geo / form
                                     ▼
   ┌──────────────────────────────────────────────────────────────────┐
   │                           Agentic layer                          │
   │  Intake → Geo-Enrichment → Validation → Triage → Enrichment      │
   │                            ↓                                     │
   │       Notification ← Cluster Detection ← Knowledge Update        │
   └─────────────┬──────────────────────────┬─────────────────────────┘
                 │                          │
                 ▼                          ▼
   ┌──────────────────────┐     ┌───────────────────────────────────┐
   │  DuckLake knowledge  │     │           MCP servers              │
   │  graph (this repo)   │ ←→  │  vectorsurv-mcp ✅                 │
   │                      │     │  nws-heatrisk-mcp ⏳               │
   │  observations,       │     │  mag-hrn-mcp ⏳                    │
   │  pathogens,          │     │  adhs-mcp ⏳                       │
   │  outbreaks,          │     │  great-az-tick-check-mcp ⏳        │
   │  resources, regions, │     │  whispers-mcp ⏳                   │
   │  tribes, ICD-10, …   │     │  inaturalist-mcp ⏳                │
   │                      │     │  211-az-mcp ⏳                     │
   └──────────────────────┘     │  knowledge-graph-mcp ⏳            │
                                └───────────────────────────────────┘
                 ▲
                 │
   ┌──────────────────────────────────────────────────────────────────┐
   │  Agency analyst / CHW / decision-maker dashboards (web)          │
   │  Backed by the same knowledge graph; recursive-CTE path queries  │
   └──────────────────────────────────────────────────────────────────┘

Why this approach

Status snapshot

Component Status
Knowledge-graph schema (core + deep seeds) ✅ shipped
MapLibre + Cytoscape viewers ✅ shipped
vectorsurv-mcp (spec-aligned to v1.0.44) ✅ shipped
nws-heatrisk-mcp ⏳ planned
mag-hrn-mcp ⏳ planned
adhs-mcp ⏳ planned
great-az-tick-check-mcp ⏳ planned
whispers-mcp, inaturalist-mcp, 211-az-mcp, knowledge-graph-mcp ⏳ planned
Intake / Triage / Enrichment / Notification agents ⏳ planned
Mobile UI (VBD + Heat) ⏳ planned
Agency dashboard ⏳ planned