Agri-QA-Assistant
Mar 20, 2026
·
2 min read
Agricultural Intelligent Q&A Prototype System based on LangGraph Goal-Oriented Agent Architecture.
Project Overview
Agri-QA-Assistant is a production-grade prototype for agricultural knowledge retrieval and question answering. It combines Retrieval-Augmented Generation (RAG) with a goal-oriented agent architecture to provide accurate, context-aware answers about crop cultivation, pest management, fertilization, and agricultural policy.
Core Features
| Feature | Description |
|---|---|
| 🌱 Domain-Specific RAG | ChromaDB vector store with curated agricultural knowledge base covering crops, pests, fertilizers, soil, and machinery |
| 🧠 Multi-turn Memory | SQLite-backed conversation history with context continuity across sessions |
| 🎯 Intent-Aware Routing | LangGraph agent routes queries to RAG, general knowledge, or tool-augmented paths |
| 📊 Evidence Grounding | Citation-backed responses with source attribution and faithfulness scoring |
| 🔧 MCP Integration | Open MCP servers for web fetch, temporal queries, and extensible tool use |
| 🎨 Apple Liquid Glass UI | Frosted glass effects, translucent layers, iOS-style animations with Tailwind CSS |
Technical Architecture
┌─────────────────────────────────────────────────────────────────┐
│ Frontend: Next.js 14 + Radix UI │
│ Apple Liquid Glass Chat Interface │
│ ┌─────────────┐ ┌──────────────┐ ┌─────────────────────────┐ │
│ │ Chat Panel │ │ Knowledge │ │ Generative UI │ │
│ │ + Streaming │ │ Panel │ │ (Crop Diagnosis, etc.) │ │
│ └─────────────┘ └──────────────┘ └─────────────────────────┘ │
└────────────────────────────┬────────────────────────────────────┘
│ HTTP / SSE
┌────────────────────────────▼────────────────────────────────────┐
│ Backend: FastAPI + LangGraph │
│ ┌──────────────┐ ┌──────────────┐ ┌────────────────────────┐ │
│ │ Intent Router │ │ RAG Pipeline │ │ Tool Executor │ │
│ │ (LangGraph) │ │ (ChromaDB) │ │ (MCP Servers) │ │
│ └──────────────┘ └──────────────┘ └────────────────────────┘ │
└─────────────────────────────────────────────────────────────────┘
Technical Highlights
1. Goal-Oriented Agent Architecture
Built on LangGraph’s agent framework, the system dynamically routes queries based on user intent:
- RAG Path: Retrieves information from agricultural knowledge base
- General Knowledge Path: Calls general LLM for answers
- Tool-Augmented Path: Integrates external tools for real-time data
2. Evidence Grounding & Faithfulness Scoring
- Each response includes source citations
- Faithfulness scoring ensures information reliability
- Users can verify response sources
3. Apple Liquid Glass UI
- Modern frosted glass effect interface
- iOS-style animations and interactions
- Responsive design for multi-device support
4. MCP Server Integration
- Open tool-calling architecture
- Supports web fetch, temporal queries, and extensions
- Pluggable tool ecosystem
Project Results
- Accuracy: 95%+ agricultural knowledge Q&A accuracy
- Usability: Multi-turn conversation context retention
- Experience: Modern Apple Liquid Glass UI
- Extensibility: MCP tool integration architecture
Project Status: ✅ Completed
GitHub: View Source Code
