BeejX: The Offline AI Ecosystem for Indian Farmers

"Digital Agriculture shouldn't stop where the Internet stops."

BeejX is an Offline-First Super App designed to empower the 400 million smallholder farmers in India who struggle with connectivity. It combines Edge AI (On-device LLMs), IoT (Soil Monitoring), and Computer Vision (Disease Detection) into a single, seamless Flutter application.

---

🚀 The Problem

While the world talks about "Cloud AI," a farmer in a remote hill village in Uttarakhand has no signal.

• Existing apps fail without 4G.
• IoT Hardware is too expensive ($500+).
• Disease Diagnosis requires sending photos to a server (which fails due to latency).
• Everything at ONE Place

💡 The Solution: BeejX

We built an operating system for the farm that works 100% Offline when needed, and syncs to the cloud when possible.

Key Features

• 🧠 Offline Brain (Samvaad): Runs Google's Gemma-2 (270M) LLM locally on the phone. Farmers can ask "How to fix yellow leaves?" in Hindi/English without internet.
• 👁️ Vaidya (Crop Doctor): Uses a custom MobileNetV2 (TFLite) model to detect diseases like Rice Blast or Wheat Rust in < 200ms using the camera.
• 📡 Bijuka (IoT Sentinel): A low-cost (< ₹2000) hardware kit (Arduino + ESP8266) that monitors Soil pH and Moisture live.
• 📒 Lekha (Smart Ledger): Digitizes farm expenses by scanning bills using OCR.

---

Tech Stack (Google Ecosystem)

This project is built almost entirely on Google Technologies:

• Mobile: Flutter (Dart) - Material 3 Design.
• Backend: Firebase (Auth, Realtime Database for IoT).
• AI (Cloud): Gemini API (for complex Query RAG).
• AI (Edge): TensorFlow Lite (Vision) + Gemma (Text via llama.cpp).
• Hardware: ESP8266 NodeMCU + Arduino Mega.

---

The Journey: Challenges I Faced

Building BeejX wasn't just about writing UI code. It was a battle against hardware constraints. Here is the real story:

1. The "Offline Brain" Nightmare

Getting an LLM to run on an Android phone was the hardest part.

• The Challenge: Integrating llama.cpp with Flutter using Dart FFI. I initially faced constant crashes because the Android NDK libraries weren't linking correctly.
• The Fix: I had to manually configure ndk.abiFilters 'armeabi-v7a', 'arm64-v8a' in build.gradle and ensure the .gguf model was quantized to q8_0 to balance speed and accuracy without blowing up the RAM.

2. Taming the IoT Latency

Connecting an Arduino to a beautiful Flutter UI sounds easy, but "Real-time" is hard.

• The Challenge: The ESP8266 would sometimes disconnect in the field, causing the App to show stale data (e.g., showing "Pump ON" when it was actually OFF).
• The Fix: I implemented a Heartbeat Mechanism in Firebase. The hardware updates a timestamp every 5 seconds. If the Flutter app sees the timestamp is > 10 seconds old, it instantly grays out the UI and shows a "Sensor Offline" warning.

3. Making it "Not Ugly"

Most Agri-apps look like boring government forms.

• The Goal: I wanted BeejX to look like a premium SaaS product.
• The Solution: I used Glassmorphism (frosted glass effects) and fl_chart for the Bijuka dashboard. Adapting these high-end visuals to perform smoothly on low-end devices required optimizing the render loop to avoid "Jank."

---

🔧 How to Run

1. Clone the Repo:

   git clone https://github.com/BhashkarFulara369/BeejX.git

2. Dependencies:

   flutter pub get

3. Model Setup:
   • Download gemma-2-270m-it.gguf (Quantized).
   • Place it in the phone's storage or assets.
4. Run:

   flutter run --release

   (Note: Connect a physical Android device. Emulators are too slow for the Offline Brain.)

---

Future Roadmap

• Voice-to-Voice (Hinglish): Fully conversational mode using OpenAI Whisper (Offline).
• LoRaWAN Support: Replacing ESP8266 Wi-Fi with LoRa for 5km+ range.
• Drone Integration: For aerial disease mapping.

---

me it's me