🌾 Kisan Alert — Smart Water, Crop & Advisory System

Voice/SMS agricultural intelligence for small & marginal farmers in Indic languages. Helps farmers pick the right crop, warns them about dry spells, and diagnoses crop disease from a photo or voice note — routed to extension officers (Rythu Seva Kendra / Krishi Vigyan Kendra) for follow-up.

Pilot regions: Maharashtra (Marathi) and Telangana (Telugu).

What it does

1. Crop recommendation — ranks crops for a farmer's field from rainfall, soil moisture, groundwater & NDVI, with plain-language reasons.
2. Dry-spell alerts — a daily job checks each field's forecast and pushes a localized irrigation advisory when a dry spell is coming.
3. Crop-health diagnosis — a photo or voice complaint → Gemini multimodal → structured diagnosis + advice; serious cases escalate to an officer.

Stack (all Google, TypeScript-only — zero Python)

Concern	Tech
Backend	TypeScript on Cloud Run (Express)
App data	Firestore (kisan-db, native)
Weather	Open-Meteo (free, no key)
Satellite NDVI / soil moisture	Earth Engine (JS client, server-side compute)
Vision/voice diagnosis	Gemini (gemini-2.5-flash)
Speech / TTS / translate	Google Cloud Speech, Text-to-Speech, Translation
Officer console	Firebase Hosting + Maps Platform
Scheduling	Cloud Scheduler

Monorepo layout

packages/
  core/       domain types + config (reads secrets/)
  advisory/   dry-spell rules engine + localized messages   [8 unit tests]
  data/       Open-Meteo weather connector (WeatherProvider)
  reco/       crop knowledge base + scoring + Earth Engine NDVI sampler
  diagnosis/  Gemini multimodal crop-health diagnosis
  lang/       Google Speech / TTS / Translation (Bhashini-swappable interface)
  db/         Firestore client + repositories (farmers, alerts, diagnoses)
apps/
  gateway/    Express on Cloud Run — webhooks, dry-spell job, reco, diagnose
  dashboard/  static officer console (Firebase Hosting + Maps)

Status — verified live

• ✅ Dry-spell engine (8 unit tests) + real Open-Meteo forecast → Marathi/Telugu message
• ✅ Smart crop reco with real signals — Earth Engine NDVI (Sentinel-2) + soil moisture (SMAP) + SoilGrids pH/soil-type + CGWB groundwater + Open-Meteo seasonal-rainfall archive, assembled per field, cached per day in Firestore. Beed → NDVI 0.14, SM 0.19, pH 7.3, GW 9 m, 1076 mm → drought-hardy crops top.
• ✅ Diagnosis → officer follow-up loop — photo stored in Cloud Storage, Gemini diagnosis with farmer/crop/field/season + prior-diagnosis context, escalation to RSK/KVK, officer claim → expert advice → auto-translated SMS back to the farmer → resolve. Officer console drives it via the gateway.
• ✅ Gemini diagnosis (voice → bollworm, Marathi advice, KVK escalation)
• ✅ Google Translate + Text-to-Speech (Marathi neural voice)
• ✅ Firestore read/write/delete on kisan-db
• ✅ Gateway end-to-end: SMS registration flow → reco → dry-spell job → diagnose
• ✅ Farmer phone simulator (apps/simulator/) drives the live gateway
• ⏳ Speech-to-Text — wired, not yet exercised with real audio

Run locally

npm install
npm run build
npm test                                   # advisory unit tests
node apps/gateway/dist/server.js           # gateway on :8080

Secrets live in secrets/ (gitignored): service-account.json + keys.txt (maps=…, gemini=…). Config is read by @kisan/core.

Try the gateway

# Register a farmer over "SMS"
curl -sX POST localhost:8080/webhook/sms -d '{"from":"+919812345678"}' -H 'content-type: application/json'
curl -sX POST localhost:8080/webhook/sms -d '{"from":"+919812345678","text":"1"}' -H 'content-type: application/json'   # Marathi
# ...name, state, crop...

curl -s localhost:8080/reco/farmer-919812345678          # crop recommendation
curl -sX POST localhost:8080/jobs/dry-spell              # run the alert loop
curl -sX POST localhost:8080/diagnose -d '{"farmerId":"farmer-919812345678","voiceTranscript":"leaves have holes"}' -H 'content-type: application/json'

Deploy

# Gateway → Cloud Run + daily dry-spell scheduler
bash deploy/deploy.sh
# Officer dashboard → Firebase Hosting (+ Firestore rules)
cp apps/dashboard/public/config.example.js apps/dashboard/public/config.js   # fill in
firebase deploy --only hosting,firestore:rules

GCP setup (one-time): enable Firestore, Translation, Text-to-Speech, Speech-to-Text, Earth Engine, Maps APIs; create the kisan-db Firestore database; register the project for Earth Engine (non-commercial).

Notes / next steps

• Telephony is mocked (MockMessageProvider) — swap for Exotel/Gupshup (DLT-registered) by implementing MessageProvider.
• Conversation sessions are in-memory — move to Firestore/Redis for multi-instance.
• lang/ is an interface; a Bhashini implementation can replace GoogleLang for better rural dialect coverage.
• Field location defaults to a district centroid at registration — capture real GPS/SMS coordinates for accurate weather/NDVI.