SmartWaste — Restaurant Waste Intelligence Platform

Real-time CV + LLM pipeline that scans commercial waste bins, classifies food and plastic waste, scores sustainability (1–4 scale), and triggers voice + email alerts when plastic thresholds are exceeded.

DataHacks 2026 Submission

---

Architecture

iOS Camera
  │  presigned PUT
  ▼
S3 (snaptrash-raw-incoming)
  │  Lambda trigger
  ▼
AWS Lambda
  ├─ Rekognition  — dedup (similarity > 0.85 → skip re-analysis)
  └─ Grok Vision  — classify food + plastic items
       │  enriched ScanRow
       ▼
  Databricks Delta (snaptrash.scans)
       │  02_aggregations + 03_prophet_forecast notebooks
       ▼
  INSIGHTS + LOCALITY_AGG tables
  (sustainability score 1–4, Prophet forecast, ZIP ranking)
       │
       ▼
FastAPI :8000  (/insights, /locality, /weekly-series, /scan/latest)
       │  Vite proxy /api
       ▼
React Dashboard  (shadcn/ui + recharts, 30s live refresh)

Voice Alerts: Vapi.ai call if plastic > 150 kg/week (locality)
Email Alerts: SMTP if plastic > 150 kg/week (dedup: 1 email/ZIP/7 days)

---

Quickstart

Prerequisites

• Python 3.11+, Node 18+
• Databricks workspace with SQL warehouse
• AWS account (S3 + Lambda + Rekognition + DynamoDB)
• xAI API key (Grok Vision)

1. Environment

cp .env.example .env
# Fill in all values — see .env.example for descriptions

Required vars:

DATABRICKS_HOST=https://your-workspace.azuredatabricks.net
DATABRICKS_TOKEN=dapi...
DATABRICKS_WAREHOUSE_ID=...
DATABRICKS_USER=you@email.com
DATABRICKS_CATALOG=workspace
S3_BUCKET=snaptrash-bins
S3_RAW_BUCKET=snaptrash-raw-incoming
XAI_API_KEY=xai-...

2. Bootstrap Databricks tables

cd scripts
python bootstrap_databricks.py
python seed_fake_scans.py   # optional: 280 synthetic scans for demo

3. Start backend

cd apps/ingestion
pip install -e ../../packages/common -e .
uvicorn snaptrash_ingestion.main:app --reload --port 8000

API docs: http://localhost:8000/docs

4. Start frontend

cd apps/frontend
npm install
npm run dev   # http://localhost:5173

Vite proxies /api/* → http://localhost:8000.

5. Run analytics (Databricks)

Upload notebooks from apps/analytics/notebooks/ to your Databricks workspace at /Users/{DATABRICKS_USER}/snaptrash/. Run manually or let pipeline_trigger auto-submit after each scan (90s cooldown).

6. Voice + email alerts (optional)

# Set in .env: VAPI_API_KEY, VAPI_ASSISTANT_ID, VAPI_PHONE_NUMBER_ID
#              DEFAULT_ALERT_PHONE, SMTP_USER, SMTP_PASS
#              ALERT_FROM_EMAIL, ALERT_TO_EMAILS

cd apps/voice-alerts
pip install -e ../../packages/common -e .
python -m snaptrash_voice_alerts.trigger

---

Endpoints

Method	Path	Description
GET	/health	Service health check
POST	/scan	Upload image (multipart) → full pipeline
GET	/upload-url	Presigned S3 PUT URL for iOS direct upload
GET	/scan/latest/{restaurant_id}	Most recent scan row
GET	/insights/{restaurant_id}	7-day aggregates + score + Prophet forecast
GET	/locality/{zip}	ZIP-level plastic + sustainability stats
GET	/weekly-series/{restaurant_id}	Day-of-week actual food kg (past 7 days)

---

Sustainability Score (1–4 scale)

Five equally-weighted signals (20% each):

Signal	Measures	Reference
S1	Food kg vs ZIP average (1% rule)	EPA MSW 2022
S2	Banned + harmful plastic penalty	CA SB-54, IARC Group 2B
S3	Recyclability rate (PET/HDPE/PP)	EPA MSW 2022
S4	Total plastic kg vs ZIP average	ZIP rolling 7d
S5	Week-over-week reduction	EPA 20% voluntary goal

score = 1.0 + (raw_0_100 / 100) × 3.0 → clamped to [1.0, 4.0]

Tiers: Thriving Forest ≥3.7 · Full Tree ≥3.4 · Growing Plant ≥3.1 · Small Sprout ≥2.8 · Seed ≥2.5 · Bare Root <2.5

---

Project Structure

apps/
  ingestion/        FastAPI service — scan ingestion + analytics routes
  analytics/        Databricks notebooks + Python aggregation scripts
  frontend/         Vite + React + shadcn/ui live dashboard
  voice-alerts/     Vapi.ai voice calls + SMTP email alerts
packages/
  common/           Shared schemas, env, table DDL, Databricks client, jobs API
infrastructure/
  lambda-detector/  AWS Lambda S3 trigger (Rekognition dedup + Grok pipeline)
scripts/
  bootstrap_databricks.py   Create all Delta tables
  seed_fake_scans.py        Generate 280 synthetic demo rows
  voice_alert_call.py       Manual alert trigger
data/
  epa_banned.json   Banned plastic polymers by state (SB-54, SB-270, NY S1185, WA SB 5022…)

---

Alert Thresholds

Alert	Threshold	Dedup
Locality plastic	>150 kg/week across ZIP	1 email/ZIP/7 days (EMAIL_ALERTS table)
Restaurant plastic	>30 kg/week per restaurant	per-run only
Rekognition dedup	similarity >0.85	per-image (DynamoDB)

---

Key Dependencies

Backend: FastAPI · Pydantic · httpx · databricks-sql-connector · boto3 · Prophet (Databricks)
Frontend: React 18 · Vite · shadcn/ui · Recharts · TanStack Query · Tailwind CSS
Infrastructure: AWS Lambda · S3 · Rekognition · DynamoDB · Databricks Delta Lake · Vapi.ai