Event material recovery operations platform. A Next.js operations dashboard plus a Flutter field-worker app that turn post-event waste into traceable, recoverable material batches.

src/                     Next.js 14 web app (ops dashboard + mobile web views)
  app/                   Routes (dashboard, batches, crews, dispatch, impact, zones, ...)
  components/            UI, layout, dashboard, batches, assistant components
  lib/                   Mock data, types, formatters, agents, assistant logic
  store/                 Zustand store (useMaterialOpsStore)
mobile_flutter/          Flutter field-worker app (scan, tasks, recovery plans)
  lib/                   models, state (provider/AppState), screens, widgets, cv
ml/                      CV model training script (build_material_model.py)

Stack: Next.js 14 (App Router), TypeScript, Tailwind CSS, Zustand, Recharts.

npm install
npm run dev        # http://localhost:3000
npm run build      # production build (type-checked)

• Ops dashboard — data-driven recovery board with zone filtering and priority/weight/value sorting; live top-strip metrics derived from batch and task data (recovered tonnage, reuse rate, active pickups, batch count).
• Batch detail & material passport — per-batch passport, grouped item breakdown, chain of custody, and priority reasoning, all driven by the batch record.
• AI agents (src/lib/agents.ts) — five deterministic agents (Material, Reuse, Repair, Logistics, Donation) that emit a shared AgentDecision schema (summary, suggested path, confidence, signals, recommendations). Surfaced on the batch detail page via AgentDecisions.
• Assistant (src/lib/assistant.ts + AssistantPanel) — a rule-based chat assistant available on every page that answers questions about batches, zones, recovery paths, crews, and overall impact using the in-app data. No external LLM required; works offline.

The field-worker app lets crews scan material, confirm/correct AI detections, group items into batches (creating material passports), and follow generated recovery plans.

• MaterialPassport, MaterialBatchSummary, RecoveryPlan models with JSON serialization.
• AppState (provider) persists passports to disk (passports.json via path_provider) and derives batch summaries and recovery plans.
• Scan screen: confirm/correct detections, assign to an existing or new batch, add a worker note, and view grouped batches.
• Task detail: per-batch recovery plan card with step progress.

Requires the Flutter SDK (not bundled). Install from https://docs.flutter.dev/get-started/install, then:

cd mobile_flutter
flutter create .          # generate platform folders (first run only)
flutter pub get
flutter analyze
flutter run

The app runs in mock detection mode until a TFLite model is present at mobile_flutter/assets/models/material_classifier.tflite. See ml/ for the training script.

ml/build_material_model.py trains a MobileNetV2-based classifier (6 classes, 224×224) and exports a .tflite model into the Flutter assets folder. Requires TensorFlow (pip install tensorflow).

• The web app uses mock data in src/lib/mockData.ts; there is no backend.
• Agents and the assistant are deterministic prototypes designed to be replaced by live services later.