AgroAssist Chatbot : RAG-Based Smart Farming Assistant

An end-to-end Retrieval-Augmented Generation (RAG) conversational AI system designed to deliver context-aware, reliable, and explainable agricultural assistance.

The AgroAssist Chatbot enables users to ask questions about plant diseases, remedies, fertilizers, and farming practices, leveraging a domain-specific knowledge base and large language models.

🚀 Features

🔍 Semantic search over agricultural knowledge base
🤖 Context-aware responses using LLaMA 3 (via Groq)
📄 Source attribution for transparency
🧠 Session-based conversational memory
🎤 Voice input (Speech-to-Text)
🔊 Voice output (Text-to-Speech)
⚡ Low-latency inference pipeline
🌐 Interactive web-based chatbot UI

🏗️ Architecture

User (Web UI)
↓
Frontend (HTML/CSS/JS + Voice)
↓
FastAPI Backend
↓
RAG Pipeline
├── Embeddings (MiniLM)
├── Vector DB (ChromaDB)
└── LLM (LLaMA 3 via Groq)
↓
Response + Sources

🔄 Workflow

1. Knowledge Base Ingestion

• Load agricultural PDFs using PyMuPDF
• Split text into chunks (RecursiveCharacterTextSplitter)
• Generate embeddings using MiniLM
• Store vectors in ChromaDB

2. Query Processing

• User sends query via frontend
• Backend receives request through FastAPI (/api/chat)

3. Semantic Retrieval

• Query is embedded
• Top-k relevant chunks retrieved from vector database

4. Context-Grounded Generation

• Retrieved context injected into prompt
• LLM generates response using:
  • Context
  • User query
  • Conversation history

5. Response Delivery

• Returns answer + source documents
• Displayed in UI (with optional voice output)

🧰 Tech Stack

AI / ML

• LangChain
• Sentence Transformers (all-MiniLM-L6-v2)
• LLaMA 3 (Groq API)

Backend

• FastAPI
• Pydantic

Vector Database

• ChromaDB

Frontend

• HTML, CSS, JavaScript

Speech

• Web Speech API (STT + TTS)

💡 Design Decisions

• RAG over Fine-Tuning → Dynamic and cost-efficient
• Context Restriction Prompting → Reduces hallucinations
• Lightweight Embeddings (MiniLM) → Fast and efficient
• Source Attribution → Improves trust
• Groq Inference → Low-latency responses

🔮 Future Improvements

• Hybrid search (semantic + keyword)
• Persistent memory (Redis / DB)
• Real-time data integration
• Response streaming
• Guardrails / moderation layer
• Multilingual support

📌 Use Case

Designed for farmers and agricultural stakeholders to:

• Diagnose plant diseases
• Get actionable remedies
• Access reliable farming knowledge
• Interact via voice