Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam December 26, 2025 Soonho Kim and Yanyan Liu CGIAR Science Program on Sustainable Farming Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 1 of 11 CGIAR Contents Executive Summary 2 Introduction 2 Data 2 AI Agent Development 3 Prototype 4 Next steps 5 Annex 5 Figure 1. System architecture diagram for AI agent 4 Figure 2. A screenshot of the application 5 CGIAR Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 2 of 11 Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam Executive Summary This activity aligns strategically with the CGIAR Sustainable Farming Science Program, specifically contributing to Area of Work 4: Plant Health and Mycotoxin Safe Crops. In response to the 2025 implementation plan's mandate for improved high-throughput, cost-effective monitoring tools, this project addresses a critical surveillance gap in Vietnam’s agricultural sector. Transboundary pests, diseases, and weeds pose significant threats to national food security and export value chains. While traditional physical surveillance methods are accurate, they remain resource-intensive and geographically constrained. Consequently, vital early warning signals, which often surface in local news and digital media prior to official scientific confirmation, are frequently missed, leading to delayed responses. To bridge this intelligence gap, the activity is developing an automated media monitoring tool powered by an AI architecture hosted on the Google Cloud Platform. Utilizing the Google Agent Development Kit and Gemini 2.5 Pro Large Language Models, the system functions as a digital analyst that automates the retrieval and synthesis of biosecurity information via Google Search. A distinguishing feature of this solution is its visual-first, designed to democratize access to complex data. By allowing users to identify threats through high-resolution images rather than requiring knowledge of scientific names, and by enabling conversation in local Vietnamese, the tool will empower a broad spectrum of stakeholders from local extension officers to national partners to access real-time, actionable insights. The foundation of this system is a comprehensive "Target Watchlist" constructed by synthesizing international data from the EPPO Global Database with Vietnam's National Technical Regulation on Phytosanitary Requirements (QCVN 01-192:2020/BNNPTNT). This dataset ensures the monitoring is economically relevant by filtering for pests affecting Vietnam's primary staple, industrial, and export fruit crops. Currently in active development, the project is moving toward a 2026 roadmap that includes rigorous ground-truthing of data with domain specialists and expanding the solution's geographic scope to India. By integrating management intelligence from trusted platforms such as Digital Green, PlantVillage, and Plantix, the agent will evolve from a detection tool into a comprehensive early warning platform for pests, diseases, and weeds. Introduction This activity is aligned with the CGIAR Sustainable Farming Science Program, contributing to Area of Work 4: Plant Health and Mycotoxin Safe Crops. As outlined in the 2025 implementation plan, the program requires the delivery of High-Level Output 4.2: an "Improved high-throughput cost-effective Pest, disease and weed management monitoring" tool. This work addresses a specific gap in Vietnam's agricultural sector, where productivity is affected by transboundary pests, diseases, and weeds such as Fall Armyworm (Spodoptera frugiperda), Coffee Leaf Rust, and invasive weed species. While traditional physical surveillance is accurate, it is often resource-intensive and limited in geographic scope. As a result, early indicators of outbreaks often appear in local news and digital media before they are officially confirmed by scientific bodies. To address this, this activity develops an automated media monitoring tool. Utilizing an AI agent that work with Large Language Models (LLMs) to identify pest, disease, and weed risks, this tool automates the retrieval of relevant information via Google Search. Crucially, the system features a multilingual conversational user interface, enabling stakeholders from local extension officers to national partners to query the data and receive results in both Vietnamese and English. By synthesizing scattered online information into accessible summaries, the system aims to provide the cost-effective and scalable monitoring capabilities required by Area of Work 4. Data The list of pests, diseases, and weeds that exist in Vietnam and the potential list from outside the country are constructed by synthesizing two authoritative biosecurity lists. First, to identify the list of pests, diseases, and weeds currently existing in Vietnam, the activity utilizes the European & Mediterranean Plant Protection Organization (EPPO) Global Database1. This international standard serves as the primary baseline for establishing the presence and distribution of established biological threats within the country. Second, to capture 1:https://gd.eppo.int/ Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 3 of 11 CGIAR a broad range of potential threats and regulated quarantine objects, this data is augmented with the National Technical Regulation on Phytosanitary Requirements for Imported Regulated Articles2(QCVN 01- 192:2020/BNNPTNT). This national standard defines the official schedule of quarantine pests (Group I) and regulated non-quarantine pests (Group II) that are strictly controlled to prevent their introduction and spread within the territory. By combining these sources, the project generates a comprehensive "Target Watchlist" that covers both known local issues and the critical quarantine risks monitored at the border. The resulting combined dataset encompasses three primary domains: Pests, Diseases, and Weeds. The Pest category monitors transboundary invaders and resistance-prone species, such as the Fall Armyworm (Spodoptera frugiperda) and Fruit Flies (Bactrocera spp.). The disease category focuses on high-impact fungal and viral pathogens, including Coffee Leaf Rust and Rice Blast, as well as emerging bacterial threats. The weed category specifically addresses invasive species and herbicide-resistant weeds that compete with staple crops, directly fulfilling the weed management monitoring component of Output 4.2. To ensure the monitoring is relevant to Vietnam's economy, these lists are filtered according to Primary Host Crops. The agent focuses its search on pests associated with Staple Crops (Rice, Maize, Cassava) for food security; Industrial Crops (Coffee, Pepper) for regional livelihoods; and Export Fruit Crops (Dragon Fruit, Durian, Mango) for international trade value. This host-centric filtering allows the system to prioritize threats that have the potential to cause significant economic damage to Vietnam’s key value chains. AI Agent Development The development of the AI agent is grounded in a robust, cloud-native architecture hosted on Google Cloud Platform (GCP), designed to handle high-throughput queries with low latency, as illustrated in the system architecture diagram shown in Figure 1. The core intelligence is powered by the Google Agent Development Kit 3(ADK) utilizing the Google Gemini 2.5 Pro Large Language Model 4(LLM). Unlike monolithic models, this system employs a multi-agent architecture, where a designated "Manager Agent" orchestrates a team of sub-agents responsible for specific execution tasks, such as collecting relevant information and filtering search results. The technical workflow begins at the frontend, hosted on Vercel5, which connects via a Cloud Load Balancer6 to the backend engine on Vertex AI7. This backend integrates a Retrieval-Augmented Generation 8(RAG) system with a vector database for context management and connects directly to the official Google Search Tool to fetch real- time data from Google News and the open web. 2 https://apps.fas.usda.gov/newgainapi/api/Report/DownloadReportByFileName?fileName=Vietnam+Iss ues+Draft+National+Technical+Regulations+for+Plant+Quarantine+Pests_Hanoi_Vietnam_03-19- 2020.pdf 3 https://google.github.io/adk-docs/ 4 https://gemini.google.com/ 5 https://vercel.com/ 6 https://cloud.google.com/load-balancing 7 https://cloud.google.com/vertex-ai 8 https://en.wikipedia.org/wiki/Retrieval-augmented_generation CGIAR Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 4 of 11 Figure 1. System architecture diagram for AI agent To address the challenge of complex agricultural taxonomy, the system features a visual-first user interface. Instead of requiring users to type specific scientific or common names, which they may not know, the interface presents images of the pests, diseases, and weeds identified in the user interface. When a user selects an image, the system automatically injects precise scientific and common names into the agent's context, ensuring high- accuracy search queries. Following this visual selection, users can interact with the agent using natural language queries in their local language, which is Vietnamese. The agent is programmed with an adaptive response engine that tailors the complexity of its answers to the user's specific profile, whether they are farmers, public sector officers, private sector stakeholders, academics, or development practitioners, ensuring the insights are strictly relevant and understandable for the intended audience. Prototype The prototype of the application “Early Warning AI Agent for Pest, Disease and Weed in Viet Nam” utilizes a user-centric design that prioritizes visual recognition to assist non-technical stakeholders in navigating complex biosecurity data. The interface is structurally divided into two distinct panels: a Visual Selection Dashboard on the left and an Interactive AI Workspace on the right side shown in the Figure 2. To overcome the barrier of complex scientific taxonomy, the application employs a "Visual-First" navigation strategy where users are presented with a gallery of high-resolution identification cards. These cards are organized into intuitive categories for Pest, Disease, and Weed, allowing users to quickly filter the biological threats relevant to their specific observations. Each identification card in the dashboard provides a snapshot of critical data to confirm the organism's identity before the user initiates a query. This includes the common name, scientific name, primary host crop, and current distribution status within Vietnam. To further aid in decision-making, each card is tagged with a color-coded threat level—such as Critical, High, or Moderate—helping extension officers and local authorities focus their attention on the most urgent biosecurity risks first. The right-hand panel functions as the active workspace for the system's generative AI capabilities. The system supports a hybrid interaction model: users can either select a specific organism from the visual list to automatically inject the precise scientific context into the agent's prompt, which helps eliminate user error, or they can manually type the common or scientific name if they are comfortable doing so. This flexible workflow triggers the agent to retrieve and display real-time insights, including relevant news articles and alerts. Furthermore, the system provides general guidance on how to prevent or reduce the impact of the pest, offering practical management advice. Following this retrieval, the conversational chat interface allows users to ask follow-up questions in natural language, such as requesting specific control strategies, Integrated Pest Management (IPM) recommendations, or details on recent outbreak locations. Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 5 of 11 CGIAR Figure 2. A screenshot of the application Next steps The Early Warning AI Agent is currently in an active development phase, with ongoing refinements being made to the core architecture and user interface to ensure stability and responsiveness. As the project moves into the 2026 implementation cycle, the work plan focuses on scaling the solution's accuracy, geographic reach, and depth of knowledge. A primary priority for this next phase is the rigorous verification of the pest, disease, and weed datasets. This process will involve close collaboration with domain specialists and national plant protection experts to ground-truth the "Target Watchlist," ensuring that the specific biological threats identified by AI remain scientifically accurate and aligned with evolving field realities. In parallel with this validation work, the project aims to expand its geographic scope beyond Vietnam to support the broader CGIAR mandate. The 2026 roadmap includes adapting the agent for deployment in India which requires a new list of pest, disease, and weed and biological databases to address the distinct agricultural biosecurity challenges of South Asia. To further enhance the utility of the tool, the agent’s knowledge base regarding management practices will be significantly enriched by integrating data from globally trusted digital extension platforms, including Digital Green, PlantVillage, and Plantix. Integrating these validated sources will enable the agent to provide users with more comprehensive, context-aware guidance on prevention and control strategies, moving the system beyond simple detection into actionable response. Annex Scientific Name Type Common Name Primary Host(s) Distribution Status Acacia mearnsii Plant/Weed Black Wattle Forestry Present, no details Acanthoscelides obtectus Insect Bean Weevil Beans, Pulses Present, no details Adoretus sinicus Insect Chinese Rose Beetle Rose, Beans, Grapes Present, no details Aleurocanthus spp. (woglumi, spiniferus) Insect Spiny Whiteflies Citrus, Mango Present, no details CGIAR Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 6 of 11 Aleurodicus spp. (dispersus, destructor) Insect Spiraling Whiteflies Coconut, Banana Present, no details Amrasca biguttula Insect Cotton Leafhopper Cotton, Okra Present, no details Anoplophora spp. (chinensis, horsfieldii) Insect Longhorn Beetles Hardwood trees Present, no details Aphis spp. (citricidus, glycines) Insect Aphids Citrus, Soybean Present, widespread (A. citricidus) Apriona spp. Insect Longhorn Beetles Mulberry, Apple Present, no details Aromia bungii Insect Red-necked Longhorn Stone fruits Present, no details Aulacaspis spp. Insect Scale Insects Mango, Sugarcane Present, no details Bactrocera spp. (dorsalis, zonata) Insect Fruit Flies Polyphagous Present, restricted distribution (B. dorsalis) Batocera spp. Insect Longhorn Beetles Mango, Fig Present, no details Bemisia tabaci Insect Silverleaf Whitefly Polyphagous Present, no details Brontispa longissima Insect Coconut Leaf Beetle Coconut Present, no details Chilo spp. (suppressalis, partellus) Insect Stem Borers Rice, Maize Present, no details Conopomorpha cramerella Insect Cocoa Pod Borer Cocoa Present, no details Cylas formicarius Insect Sweet Potato Weevil Sweet Potato Present, no details Diaphorina citri Insect Asian Citrus Psyllid Citrus Present, restricted distribution Eudocima fullonia Insect Fruit-piercing Moth Citrus, Fruit Present, widespread Frankliniella occidentalis Insect Western Flower Thrips Polyphagous Present, no details Halyomorpha halys Insect Brown Marmorated Stink Bug Polyphagous Present, no details Helicoverpa armigera Insect Cotton Bollworm Cotton, Maize Present, no details Hypothenemus hampei Insect Coffee Berry Borer Coffee Present, no details Leptocorisa spp. Insect Rice Ear Bugs Rice Present, no details Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 7 of 11 CGIAR Liriomyza spp. (sativae, trifolii) Insect Leafminers Vegetables Present, widespread (L. sativae) Maconellicoccus hirsutus Insect Pink Hibiscus Mealybug Hibiscus, Grape Present, no details Nilaparvata lugens Insect Brown Planthopper Rice Present, no details Oryctes rhinoceros Insect Rhinoceros Beetle Coconut, Oil Palm Present, no details Phenacoccus spp. (manihoti, solenopsis) Insect Mealybugs Cassava, Cotton Present, no details Phthorimaea operculella Insect Potato Tuber Moth Potato, Tobacco Present, no details Rhynchophorus ferrugineus Insect Red Palm Weevil Coconut, Palms Present, no details Scirtothrips dorsalis Insect Chilli Thrips Chilli, Tea Present, no details Spodoptera spp. (frugiperda, litura) Insect Armyworms Maize, Rice Present, restricted distribution (S. exempta) Sternochetus mangiferae Insect Mango Seed Weevil Mango Present, no details Tetranychus spp. (evansi, kanzawai) Mite Spider Mites Tomato, Tea Present, no details Thrips palmi Insect Melon Thrips Cucurbits, Eggplant Present, no details Xyleborus / Xylosandrus / Euwallacea Insect Ambrosia Beetles Avocado, Trees Present, no details 'Candidatus Liberibacter asiaticus' Bacteria Citrus Greening (HLB) Citrus Present, restricted distribution Acidovorax avenae Bacteria Bacterial Stripe Rice, Corn Present, restricted distribution Pantoea stewartii Bacteria Stewart’s Wilt Maize Present, no details Pseudomonas syringae pv. tabaci Bacteria Wildfire Tobacco Present, no details Ralstonia pseudosolanacearum Bacteria Bacterial Wilt Solanaceae Present, widespread Ralstonia solanacearum (Race 2 / Complex) Bacteria Bacterial Wilt / Moko Banana, Potato Present, no details CGIAR Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 8 of 11 Xanthomonas albilineans Bacteria Leaf Scald Sugarcane Present, no details Xanthomonas citri pv. citri Bacteria Citrus Canker Citrus Present, widespread Xanthomonas oryzae (pv. oryzae) Bacteria Bacterial Blight Rice Present, restricted distribution Xanthomonas phaseoli Bacteria Common Bacterial Blight Beans Present, no details Alternaria brassicae Fungus Dark Leaf Spot Brassicas Present, no details Ceratocystis fimbriata Fungus Wilt / Canker Sweet Potato, Coffee Present, restricted distribution Colletotrichum / Glomerella Fungus Anthracnose Various fruits Not listed in distribution data Exobasidium vexans Fungus Blister Blight Tea Present, no details Fusarium oxysporum f. sp. cubense Fungus Panama Disease Banana Present, no details Harringtonia lauricola Fungus Laurel Wilt Avocado Present, no details Hemileia vastatrix Fungus Coffee Leaf Rust Coffee Present, no details Heterobasidion annosum Fungus Root Rot Conifers Present, widespread Necator salmonicolor Fungus Pink Disease Citrus, Rubber Present, no details Neopestalotiopsis rosae Fungus Crown Rot Strawberry Present, no details Oidium heveae Fungus Powdery Mildew Rubber Present, no details Peronosclerospora spp. Oomycete Downy Mildew Maize, Sorghum Present, no details Phakopsora pachyrhizi Fungus Soybean Rust Soybean Present, no details Phytophthora cinnamomi Oomycete Root Rot Polyphagous Present, widespread Phytophthora ramorum Oomycete Sudden Oak Death Oaks Present, no details Pseudocercospora fijiensis Fungus Black Sigatoka Banana Present, no details Puccinia spp. (horiana, kuehnii) Fungus Rusts Chrysanthemum Present, no details Pyricularia oryzae Fungus Rice Blast Rice Present, no details Teratosphaeria spp. Fungus Leaf Blight Eucalyptus Present, no details Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 9 of 11 CGIAR Sporisorium scitamineum Fungus Smut Sugarcane Present, widespread Aphelenchoides besseyi Nematode White Tip Nematode Rice Present, no details Ditylenchus angustus Nematode Rice Stem Nematode Rice Present, restricted distribution Hirschmanniella spp. Nematode Rice Root Nematode Rice Present, no details Meloidogyne spp. Nematode Root-Knot Nematode Rice, Guava Present, no details Pratylenchus spp. (coffeae) Nematode Lesion Nematode Coffee, Banana Present, widespread (P. coffeae) Rotylenchulus reniformis Nematode Reniform Nematode Cotton, Pineapple Present, widespread Tylenchulus semipenetrans Nematode Citrus Nematode Citrus Present, widespread Babuvirus musae Virus Banana Bunchy Top Banana Present, no details Closterovirus tristezae Virus Citrus Tristeza Citrus Present, no details Orthotospovirus meloflavi Virus Melon Yellow Spot Melons Present, no details Pospiviroid exocortis Viroid Citrus Exocortis Citrus Present, no details Pospiviroid fusituberis Viroid Potato Spindle Tuber Potato Present, few occurrences Potyvirus batataplumei Virus SP Feathery Mottle Sweet Potato Present, no details Sugarcane chlorotic streak agent Unknown Chlorotic Streak Sugarcane Present, no details Tenuivirus oryzabrevis Virus Rice Grassy Stunt Rice Present, restricted distribution Tungrovirus oryzae Virus Rice Tungro Rice Present, no details Ageratina adenophora Weed Crofton Weed Pastures Present, no details Alternanthera philoxeroides Weed Alligator Weed Waterways Present, no details Arundo donax Weed Giant Reed Riparian areas Present, no details Eichhornia / Pontederia crassipes Weed Water Hyacinth Waterways Present, no details CGIAR Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 10 of 11 Hydrilla verticillata Weed Hydrilla Waterways Present, restricted distribution Mimosa pigra Weed Giant Sensitive Plant Wetlands Present, restricted distribution Parthenium hysterophorus Weed Parthenium Weed Pastures Present, no details Striga asiatica Weed Witchweed Maize Present, no details Lissachatina fulica Snail Giant African Snail Polyphagous Present, restricted distribution Pomacea spp. (canaliculata) Snail Golden Apple Snail Rice Present, no details Early Warning AI Agent for Pest, Disease, and Weed in Viet Nam | Page 11 of 11 CGIAR Acknowledgements The CGIAR Sustainable Science Program forms a part of CGIAR’s new Research Portfolio, addressing key challenges in agri-food systems by fostering efficient production of nutritious foods and safeguarding the environment to create fair employment opportunities, as we simultaneously tackle climate change, soil degradation, pests, diseases, and desertification. We would like to thank all funders who supported this research through their contributions to the CGIAR Trust Fund: https://www.cgiar.org/funders/ About CGIAR Sustainable Science Program Report This research was conducted as part of the CGIAR Sustainable Farming Science Program. This research is being implemented by CGIAR researchers from International Food Policy Research Institute in close partnership with IFPRI’s Food Security Portal. CGIAR is a global research partnership for a food-secure future. Its science is carried out by 15 Research Centers in close collaboration with hundreds of global partners. www.cgiar.org Executive Summary Introduction Data AI Agent Development Prototype Next steps Annex