AgriSense：云原生农业 AI 智能体，融合作物病害识别与市场预测

AgriSense is a Kubernetes-native multi-service AI agent system designed for agricultural scenarios, integrating crop disease detection, market intelligence analysis, and price prediction functions. It adopts a modern cloud-native architecture, combining KEDA auto-scaling, Azure Foundry IQ, HuggingFace plant disease recognition models, and Meta Prophet time-series prediction, demonstrating the practical application value of AI Agents in agricultural digital transformation.

Basic Information:

• Original Author/Maintainer: ApurboBarua17
• Source Platform: GitHub
• Original Link: https://github.com/ApurboBarua17/agrisense
• Release Time: 2026-06-09
• Competition Background: Agents League 2026, Reasoning Agents Track

Agriculture is the cornerstone of human civilization, but traditional agriculture faces many challenges: crop yield reduction due to pests and diseases, income uncertainty caused by market price fluctuations, and difficulty in accessing agricultural technical knowledge. With the development of artificial intelligence technology, applying these advanced technologies to agricultural scenarios has become an effective way to solve these problems.

AgriSense was born in this context, representing a new idea: integrating multi-modal AI capabilities into a unified agent system to provide end-to-end intelligent decision support for farmers and agricultural practitioners. The project stood out in the Reasoning Agents track of Agents League 2026, demonstrating the practical application potential of AI Agents in vertical fields.

Kubernetes Native Design

AgriSense uses Kubernetes as the basic operating platform, which brings significant advantages:

• Elastic Scaling: Using KEDA (Kubernetes Event-Driven Autoscaling) to achieve load-based auto-scaling, increasing instances automatically during peak periods and reducing resources during off-peak periods
• High Availability: Ensuring continuous service availability through Kubernetes health checks and self-healing mechanisms
• Microservice Decoupling: Splitting different functional modules into independent services for easy independent development, deployment, and maintenance
• Cloud Agnosticism: Can run on any platform supporting Kubernetes, including public clouds, private clouds, and edge environments

Multi-Service Collaboration Architecture

The system consists of multiple specialized services, each responsible for specific functional areas:

1. Crop Disease Detection Service: Uses computer vision technology to identify disease symptoms on crop leaves, based on HuggingFace open-source models, supports multiple crops, provides real-time analysis and confidence scores.
2. Market Intelligence Service: Collects and analyzes agricultural product market data, integrates multi-source data, uses Azure Foundry IQ for intelligent processing, provides trend analysis and risk warnings.
3. Price Prediction Service: Uses Meta Prophet model for future price trend prediction, supports short/medium/long-term forecasts, provides uncertainty quantification and continuous learning.
4. Reasoning Agent Service: Acts as the system's 'brain', coordinates services, understands user needs via natural language interaction, supports multi-round dialogue, tool calling, and explainable output.

KEDA Auto-Scaling Strategy

AgriSense makes full use of KEDA's capabilities, designing refined scaling strategies:

• Queue length-based scaling: Automatically increases the number of disease detection service instances when image recognition requests pile up
• Time pattern-based pre-scaling: Pre-increases resources during sowing and harvest seasons considering agricultural seasonality
• Cost optimization: Automatically reduces to the minimum number of instances at night and slack periods to lower operating costs

Azure Foundry IQ Integration

AgriSense uses Azure AI Foundry's IQ function to achieve:

• Intelligent document parsing: Automatically extracts key information from unstructured agricultural reports and news
• Multi-language support: Can process agricultural information from different countries and regions
• Knowledge graph construction: Organizes extracted information into structured knowledge graphs to support complex queries

Prophet Model Adaptation for Agriculture

AgriSense optimizes the Prophet model for agricultural price prediction challenges:

• Holiday effect modeling: Considers the impact of traditional festivals like Spring Festival and harvest festivals on demand
• Weather data fusion: Incorporates weather forecasts as external regression variables into the model
• Outlier handling: Identifies and properly handles price anomalies caused by extreme weather or emergencies

Scenario 1: Rapid Disease Diagnosis

When a farmer finds abnormal spots on crop leaves in the field, he immediately takes a photo with his mobile phone and uploads it to AgriSense. The system identifies it as early symptoms of rice blast within seconds and provides prevention suggestions: recommended pesticides, best application time, and preventive measures.

Scenario 2: Sales Timing Decision

A large grower plans to sell this season's corn but is unsure whether to sell now or wait for a better price. He asks AgriSense about the price trend in the next two weeks. The system comprehensively analyzes current inventory data, market demand forecasts, and weather impacts, suggesting he sell in a week, which is expected to get an 8% price increase.

Scenario3: Planting Planning Assistance

An agricultural cooperative leader is planning next season's crop planting. Through dialogue with AgriSense, he learns that according to current market trend forecasts, soybean prices are expected to rise while wheat prices may fall. Based on this information, he decides to adjust the planting ratio and increase the soybean planting area.

Cloud-Native Architecture Advantages

• Edge Deployment Capability: Can run on lightweight Kubernetes distributions like K3s/MicroK8s, has offline capabilities for core functions, and supports edge-cloud collaboration (edge nodes handle real-time requests, cloud is responsible for model training and big data processing), suitable for rural areas with imperfect network infrastructure.
• Multi-Tenant Support: Provides namespace isolation for data of different farmers/cooperatives, resource quota management, and customized configuration for each tenant.

Comparison with Other Agricultural AI Solutions

Feature	AgriSense	Traditional Agricultural App	General AI Assistant
Disease Recognition	Professional model, multi-crop support	Limited or dependent on manual work	General capability, low accuracy
Price Prediction	Time-series model, agricultural adaptation	Usually no such function	No special optimization
Cloud-Native Architecture	Kubernetes + KEDA	Mostly monolithic applications	Varies
Auto-Scaling	Natively supported	Usually none	Depends on implementation
Interpretability	Transparent reasoning process	Black-box decision	Varies

Project Significance & Industry Impact

AgriSense's significance lies not only in technical implementation but also in demonstrating the huge potential of AI Agents in traditional agriculture:

• Lowering Technical Threshold: Farmers can get professional advice through natural language dialogue without understanding complex AI models, greatly reducing the threshold for using agricultural intelligent technology.
• Data-Driven Decision Making: Traditional agricultural decisions often rely on experience inheritance, while AgriSense provides quantitative decision support based on big data and machine learning, helping farmers make more scientific decisions.
• Replicable Architecture Pattern: The project's architecture design (Kubernetes native, multi-service collaboration, event-driven scaling) provides a replicable reference pattern for AI Agent applications in other vertical fields.

Future Development Directions

Based on the current architecture, AgriSense can evolve in the following directions:

• IoT Integration: Access soil sensors, weather stations and other IoT devices to achieve more accurate data collection.
• Drone Collaboration: Integrate with agricultural drone systems to achieve large-scale crop health monitoring.
• Blockchain Traceability: Combine blockchain technology to establish a full traceability system from planting to sales.
• Multi-Modal Expansion: In addition to image recognition, add processing capabilities for video and audio (such as农机故障 sounds)