EcoAgriGrid

EcoAgriGrid Login Page — Supports Email, Google OAuth, and Guest Mode with English/Hindi/Marathi language switcher.
Central Hub — Choose between Traditional Farming AI, Autonomous Hydroponics Dashboard, or Farmer Support Hub
5 AI Modules: Crop Recommendation (LightGBM), Fertilizer Optimizer, Disease Detection (CNN), Irrigation Advisory & Market Prediction
LangGraph AI Agent live dashboard — real-time pH, EC, temperature & humidity from ESP32 sensors with autonomous action logs
Farmer Support Hub — P2P Direct Aid, 1000+ Government Scheme Finder, and NGO Partner Directory in one place

Inspiration

India has 140 million farmers, but less than 30% have access to expert agricultural guidance. We watched farmers lose crops to preventable diseases, wrong fertilizers, and poor market timing — not because solutions don't exist, but because they're inaccessible. Hydroponics farmers face an even harder challenge: one missed pH reading can destroy an entire crop overnight. We built EcoAgriGrid to change that — putting expert-level AI in every farmer's pocket, and autonomous AI agents on every farm.

What It Does

EcoAgriGrid is a full-stack, production-deployed AI smart farming platform with three pillars:

1) Autonomous Hydroponics AI (LangGraph Multi-Agent System) : A fully autonomous IoT farm management system powered by LangGraph. ESP32 sensors stream pH, EC, temperature, and humidity to Firebase every 60 seconds. A Zero-LLM supervisor node routes to specialist AI agents only when a violation is detected — Climate Agent, pH/Nutrient Agent, Irrigation Agent, or Safety Agent. The AI writes hardware commands back to Firebase; the ESP32 executes them physically (pump on/off, fan speed, valve control) and acknowledges back. The system runs 24/7 with no human intervention.

2) AI Decision Suite for Traditional Farmers (5 ML Models) :

Crop Recommendation : LightGBM model with GPS-based agro-climatic zone detection → Top 3 crops with confidence %.
Disease Detection : TensorFlow MobileNet CNN → 38 disease conditions across 14 crop species in <2 seconds.
Fertilizer Optimizer : Cosine similarity + NPK deficit engine → exact kg/hectare dosage and cost in ₹.
Smart Irrigation Advisory : Groq LLaMA-3.3-70B (→ GPT-4o fallback) → multilingual, weather-integrated scheduling.
Market Intelligence : Statistical seasonal model → 12-month price history + 30-day predictions + MSP comparison.

3) Farmer Support Hub :

P2P verified crowdfunding
100+ government scheme finder
NGO directory

How We Built It

Frontend : Vanilla HTML/JS SPA deployed on Firebase Hosting with full EN/HI/MR localization.
Backend : Python 3.10 + Flask 3.1.2 on Azure Container Apps (Docker/Gunicorn).
ML Pipeline : LightGBM + scikit-learn (crop), TensorFlow 2.17.0 MobileNet (disease), cosine similarity (fertilizer).
AI Agents : LangGraph orchestrator + LangChain tools + Groq LLaMA-3.3-70B + gpt-4o.
IoT : ESP32 microcontrollers ↔ Firebase Realtime Database ↔ LangGraph agent loop.
Database : Firebase Firestore (users, P2P), Firebase Realtime DB (IoT), Firebase Auth.

Challenges We Ran Into

Token efficiency at scale : Designing the LangGraph supervisor to use pure Python routing (zero LLM cost) while still triggering specialist LLM agents only when needed.
ESP32 ↔ Firebase ↔ AI synchronization : Ensuring hardware commands are acknowledged and failures are safely handled without race conditions.
Multi-language ML output : Making LLMs generate advisories directly in the farmer's selected language (not translate after the fact).
Cold-start latency : Solved with background-thread model pre-loading so the first API request is instant.

Accomplishments We're Proud Of

Real IoT hardware - not simulated. Physical ESP32 nodes with real sensors controlling real actuators
Production deployed - ecoagrigrid.web.app is live today
38 disease classes across 14 crop species with <2s inference
Zero-LLM supervisor - intelligent automation without burning API tokens on every cycle
Multi-tier AI fallback - Groq → OpenAI → Static. Farmers always get a response, even if APIs are down

What We Learned

Building a truly autonomous AI system — not just a chatbot — requires engineering discipline: knowing when NOT to call an LLM is as important as knowing when to. We also learned that real-world impact requires accessibility: multilingual support and a zero-install browser-based approach were non-negotiable for Indian farmers.