🌾 Fasal Mitra (फसल मित्र)

Fasal Mitra is an AI-powered, multilingual agriculture assistant designed to support Indian farmers. By combining Retrieval-Augmented Generation (RAG) for verified agricultural practices with Agentic Web Search for real-time government schemes, it provides reliable, actionable, and localized advice step-by-step.

🔗 Try the Live App on Databricks: https://fmitra-7474650359868011.aws.databricksapps.com/

---

🚀 Features

• 🗣️ Multilingual Support: Detects and responds natively in Hindi, Bengali, Marathi, Telugu, English, and other Indian languages.
• 📚 Local Knowledge Base (RAG): Retrieves farming guidelines, crop practices, pest management, and soil health advice from a robust FAISS vector database.
• 🌐 Live Government Schemes: Uses integrated web search to fetch the latest Indian government subsidies, PM schemes, and crop insurances.
• 🤖 Agentic Routing: Intelligently decides when to search local storage vs. the web based on the farmer's query using the ReAct strategy.

---

🏗️ System Architecture

The application is orchestrated using LangGraph, employing a dynamic agent that reasons through the user's intent to select the appropriate tools.

---

🛠️ Technology Stack

Databricks Ecosystem:

• Databricks Apps: For scalable hosting and deployment of the Streamlit application.
• Databricks Notebooks: Used for development, data preprocessing, and vector store initialization.
• Delta Lake & PySpark: For robust data staging, storage, and processing (Delta tables and Spark DataFrames).

Open Source Models:

• LLM Engine: Open-source models hosted via HuggingFace Inference API (deepseek-ai/DeepSeek-R1:fastest and openai/gpt-oss-120b:fastest).
• Embeddings: sentence-transformers/all-MiniLM-L6-v2 for generating embeddings to power the RAG pipeline.

Application Framework:

• Agent Framework: LangGraph, LangChain
• Vector Database: FAISS
• Search Tool: DuckDuckGo Search API
• Frontend / UI: Streamlit (via app.py / dashboard.py)

---

⚙️ Setup & Installation

1. Clone the repository & Install Dependencies:

git clone https://github.com/princeiiti/kisan-helpbot
cd final
pip install -r streamlit-hello-world-app/requirements.txt

2. Environment Variables: Set your Hugging Face API key in your environment to enable LLM inference:

export HF_TOKEN="your_huggingface_token_here"

3. Initialize Database (first-time only): Run or step through the Jupyter Notebook New Notebook 2026-04-18 02_36_08.ipynb to construct the FAISS vector database from your text data (rag_short.txt).

4. Run the Streamlit Application:

cd streamlit-hello-world-app
streamlit run app.py

---

📊 RAG Data Pipeline

---

💡 Usage Example

User Input (Telugu):

"గోధుమ పంటకు నీటిపారుదల చేయడానికి ఉత్తమ సమయం ఏమిటి మరియు ప్రభుత్వ పథకం ఏదైనా ఉందా?" (What is the best time to irrigate the wheat crop, and is there any government scheme?)

Agent Execution Output:

1. Recognizes intent: Crop practice + Scheme inquiry.
2. Triggers search_rag("wheat crop irrigation schedule").
3. Triggers search_schemes("wheat crop farming schemes subsidies").
4. Synthesizes a unified response purely in Telugu, providing detailed irrigation steps and mentioning relevant PMKSY/insurance schemes.