New Breakthrough in Agricultural Technology: Knowledge Graph-Based Grape Cultivation AI Assistant Empowers Indian Farmers

Grape-Mind AI is an open-source AI agricultural assistant system that combines knowledge graph, vector search, and generative AI technologies to provide multilingual agronomic expert advice to Indian grape growers. This system effectively addresses the issues of Indian farmers' lack of professional knowledge and limited technical dissemination channels, supports multiple local languages, lowers the threshold for accessing technology, and facilitates precision agricultural consulting.

As a major agricultural country in the world, India's grape growers face problems such as lack of professional knowledge and limited technical dissemination channels. Traditional agricultural technology promotion is inefficient and difficult to meet individual needs. The Grape-Mind AI project emerged to provide professional agronomic advice and support multiple local languages such as Hindi and Marathi, lowering the language barrier for accessing technology.

Hybrid Retrieval Mechanism

• Structured graph data retrieval: Uses Neo4j to store entity relationships such as grape varieties, diseases, and treatment methods, enabling quick location of accurate factual information
• Unstructured document retrieval: Stores agricultural PDF embedding vectors via ChromaDB to enable semantic search

System Workflow

1. Entity extraction: Uses Gemini 2.5 Flash to extract key entities from queries
2. Graph retrieval: Traverses Neo4j knowledge graph relationships to obtain facts
3. Vector retrieval: Searches relevant document fragments in ChromaDB
4. Answer generation: Merges information and translates it into the user's chosen language

Knowledge Graph Design

Adopts a three-node, two-relationship structure: (:Variety)─[:AFFECTS]─►(:Disease)─[:TREATED_BY]─►(:Treatment), e.g., Chardonnay → Powdery Mildew → Sulfur Fungicide

Language Coverage

Supports English, Hindi, Marathi, Kannada, Telugu

Localization Considerations

• Uses agricultural knowledge suitable for local climate
• Adapts to local crop varieties and pest/disease conditions
• Provides advice aligned with local agricultural practices

• Pest and Disease Diagnosis: Describes symptoms (image recognition support planned in the future) to identify pests/diseases and provide prevention and control methods
• Variety Selection: Recommends suitable varieties based on climate and market demand
• Fertilization Management: Provides scientific fertilization advice based on soil conditions and growth stages
• Treatment Plan: Provides pesticides, application timing, and precautions for specific pests/diseases

Innovations

• Combines knowledge graph with RAG technology, balancing factual accuracy and contextual richness
• Multilingual AI assistant tailored to Indian agricultural needs, filling gaps in the field
• Open-source project with high replicability, easy to customize for other regions

Future Directions

• Expand knowledge graph content
• Support image upload for visual disease diagnosis
• Integrate weather API to provide location-based personalized advice
• Support voice input
• Export chat history as PDF reports

1. Multilingual support can be applied to agricultural scenarios in different regions of China
2. The combination model of knowledge graph and generative AI can be extended to other crop fields
3. Need to emphasize localized custom development
4. Open-source ecosystem promotes technology dissemination and improvement