PaperlayerView exact revision on arXiv

Source-linked AI summary

AgroAskAI: A Multi-Agentic AI Framework for Supporting Smallholder Farmers' Enquiries Globally

Nadine Angela Cantonjos, Arpita Biswas

arXiv:2512.14910v1cs.AIcs.CY

TL;DR

Climate-related risks create difficult adaptation decisions for smallholder farmers, and access to relevant climate information is limited in some settings. AgroAskAI combines specialized agents, real-time resources, and review to support these decisions; evaluations report more actionable, localized guidance than comparison systems. The paper concludes that agentic AI may support sustainable and accountable agricultural adaptation, while noting data, language-model, and deployment constraints.

  • Problem

    Smallholder farmers face climate-related agricultural risks, and some lack access to climate information needed for adaptation.

    Source: §1 · p. 1

  • Method

    AgroAskAI coordinates role-specialized agents with real-time external resources and reviewer critique to produce context-aware agricultural decision support.

    Source: §1 · p. 2

  • Results

    AgroAskAI produced actionable, context-aware responses and outperformed ChatGPT and CROPWAT across multiple dimensions in the reported evaluation.

    Source: §4 · pp. 6–7

  • Takeaways & Limitations

    The authors conclude that agentic AI can contribute to sustainable, scalable, and accountable climate-adaptation strategies for agriculture.

    Source: §5 · p. 7

    Performance depends on external-data quality and scope, multilingual support relies on OpenAI and may need dialect fine-tuning, and low-connectivity deployment remains challenging.

    Source: §5 · p. 7

Abstract

from arXiv · show

Agricultural regions in rural areas face damage from climate-related risks, including droughts, heavy rainfall, and shifting weather patterns. Prior research calls for adaptive risk-management solutions and decision-making strategies. To this end, artificial intelligence (AI), particularly agentic AI, offers a promising path forward. Agentic AI systems consist of autonomous, specialized agents capable of solving complex, dynamic tasks. While past systems have relied on single-agent models or have used multi-agent frameworks only for static functions, there is a growing need for architectures that support dynamic collaborative reasoning and context-aware outputs. To bridge this gap, we present AgroAskAI, a multi-agent reasoning system for climate adaptation decision support in agriculture, with a focus on vulnerable rural communities. AgroAskAI features a modular, role-specialized architecture that uses a chain-of-responsibility approach to coordinate autonomous agents, integrating real-time tools and datasets. The system has built-in governance mechanisms that mitigate hallucination and enable internal feedback for coherent, locally relevant strategies. The system also supports multilingual interactions, making it accessible to non-English-speaking farmers. Experiments on common agricultural queries related to climate adaptation show that, with additional tools and prompt refinement, AgroAskAI delivers more actionable, grounded, and inclusive outputs. Our experimental results highlight the potential of agentic AI for sustainable and accountable decision support in climate adaptation for agriculture.

1 Introduction

Climate risks threaten smallholder farming, while local adaptation requires decision support that can respond to varied and changing conditions. AgroAskAI is proposed as a role-specialized agentic system for that need.

1 Introduction

  • Climate-related risks, including erratic weather, depleted soil health, and shifting rainfall, have harmed vulnerable farmers and agriculture.In arid and semi-arid Kenya, 31% of smallholder farmers lack access to climate-related information.Source: §1 · p. 1
  • Localized studies do not meet the need for scalable adaptation support as climate patterns and extreme weather vary across regions.Smallholder farmers are described as particularly vulnerable to climate disruptions.Source: §1 · p. 1
  • Adaptation advice must be context-specific and locally adaptable because one-size-fits-all solutions risk ineffectiveness and maladaptation.The passage identifies resilient crop varieties, improved cultivation practices, and timely climate information as adaptation measures.Source: §1 · p. 1
  • Agentic AI can reason through multi-step objectives, integrate external data, and execute decision processes autonomously.The passage describes its potential to generate tailored, context-aware agricultural solutions.Source: §1 · p. 2
  • AgroAskAI combines role-specialized agents, real-time resources, and reviewer critique to support traceable, context-aware agricultural decisions.It also uses reasoning logs, decision checkpoints, and external validation to promote auditability.Source: §1 · p. 2

2 Related Work

Prior agricultural AI includes sensor-based systems and conversational tools with differing capabilities. AgroAskAI is presented as a globally oriented, multi-agent alternative with internal governance and real-time information.

2 Related Work

  • This work focuses on conversational support for farmers, rather than AI systems that automate sensor-based soil and crop monitoring.The passage states sensor monitoring systems fall outside its scope.Source: §2 · p. 2
  • Agrifriendly uses a single agent for crop-maintenance queries, which the authors say limits adaptability and lacks internal critique or modular reasoning.Farmer.Chat uses RAG for informative multilingual responses and relies on user-feedback loops and local-partner data.Source: §2 · p. 2
  • AgriBuddy combines specialized agents for Bangladeshi farmers with Bangla, English, and hybrid-language interaction and a rice-disease vision model.The agents include query handling, memory, and expert advice.Source: §2 · p. 2
  • AgroAskAI is distinguished by internal governance and access to trusted weather data and curated user-contributed documents.The authors describe these features as supporting reliable, transparent, and adaptable climate-adaptive solutions.Source: §2 · p. 2

3 Design and Methodology

AgroAskAI organizes specialized agents in a logical workflow coordinated by an Agent Manager. User input is localized, relevant agents are selected, and responses can be reviewed and revised before delivery.

3 Design and Methodology

  • The framework assigns distinct tasks to specialized agents connected through a structured workflow.The section describes agent roles and information exchange between agent pairs.Source: §3 · p. 2
  • The Prompt Agent handles user interaction and language continuity, while the Parsing Agent extracts intent, location, and timeframe and seeks clarification when needed.The parsing agent passes structured data and the query to the Agent Manager.Source: §3 · p. 2
  • The Agent Manager selects relevant agents, aggregates information for the Solution Agent, and ends the response cycle after a satisfactory answer.It typically finalizes after Reviewer Agent review.Source: §3 · pp. 2–3
  • Weather agents retrieve location-specific forecast data or historical climate records using external services and APIs.The forecast agent uses OpenCage and OpenWeather; the history agent uses OpenCage and NASA POWER.Source: §3 · p. 3
  • The Solution Agent generates actionable agricultural responses, and the Reviewer Agent checks clarity, technical soundness, fairness, and user-intent alignment.The Reviewer Agent can consult external sources and trigger another agent-selection and response cycle when an answer is incomplete or suboptimal.Source: §3 · p. 3

Model Specification

AgroAskAI uses a language-model-based multi-agent system that retrieves external data and records agent interactions. A sample exchange illustrates query routing and iterative review before an answer is delivered.

Model Specification

  • AgroAskAI uses Microsoft Semantic Kernel and OpenAI GPT-4 to process natural-language queries, with conversation-flow logging for traceability and future improvements.The architecture is described as drawing on perception, goal decomposition, and safety guardrails.Source: Model Specification · p. 3
  • RAG connects the system to external APIs for historical climate data, forecasts, and geocoding, while the Agent Manager activates agents based on user input.The passage identifies NASA POWER, OpenWeather, and OpenCage as example sources.Source: Model Specification · pp. 3–4
  • In the sample pest-control query, the Agent Manager routes directly to the Solution Agent, then reviewer feedback prompts a more specific answer before approval.Weather agents are not activated because the query does not require weather data.Source: Model Specification · p. 4, Figure 2 · p. 4

4 Experimental Results

Across queries on climate-adaptation decisions, AgroAskAI produced localized, actionable guidance and outperformed ChatGPT and CROPWAT across multiple dimensions. The evaluation also examined agent coordination, review, missing-detail handling, and multilingual interaction.

Comparison with Other Models

  • For Kitui water queries, AgroAskAI detailed water-harvesting systems and their application, while ChatGPT briefly mentioned zai pits and mulching and CROPWAT focused on irrigation metrics.All three systems gave average rainfall; AgroAskAI and ChatGPT also included dry spells and heat, while only AgroAskAI supplied temperature data (21–27°C).Source: §4 · p. 5
  • AgroAskAI offered livestock, soil-conservation, and farm-size-specific guidance, while CROPWAT lacked livestock and soil advice and ChatGPT was often less detailed or general.ChatGPT named support programs that AgroAskAI and CROPWAT did not, while both ChatGPT and AgroAskAI supported more languages than CROPWAT.Source: §4 · pp. 5–6

Importance of Agent Manager

  • The Agent Manager selected the needed specialist agents, including using only the Solution Agent when weather agents were unnecessary, and terminated conversations after a suitable solution.The examples also show queries handled by the Weather History Agent alone and the Weather Forecast Agent alone.Source: §4 · p. 6

Significance of the Reviewer Agent

  • Compared with a version without the Reviewer Agent, AgroAskAI’s responses were reported as more factually accurate and aligned with user intent, avoiding irrelevant adaptations in a drought-crop query.The passage reports that such hallucinations were absent from AgroAskAI’s reviewed responses.Source: §4 · p. 6

Handling Missing Details

  • The Parse Agent detected missing location or timestamp information when relevant, while the Prompt Agent sought clarification and the system flagged out-of-scope intent.In the example, the user supplied Kitui, Kenya after being asked for a location.Source: §4 · p. 6

Multi-lingual Capabilities

  • AgroAskAI’s Prompt Agent identified a Swahili query and responded in Swahili, with weather information presented in that language.The example response includes temperature, rainfall, humidity, wind, UV, and sunrise and sunset information.Source: §4 · pp. 6–7

Comparison with Other Models

  • AgroAskAI consistently delivered actionable, context-aware responses, outperforming ChatGPT and CROPWAT across multiple dimensions.Its step-by-step recommendations combined localized context with practical instructions for field use.Source: Table 1 · p. 5, §4 · pp. 6–7

5 Conclusion and Discussion

The paper presents AgroAskAI as a multi-agent system for localized, actionable climate-adaptation guidance and reports stronger performance than static tools and generic single-agent models. It also identifies data, language-model, and connectivity constraints, and calls for farmer-centered user testing.

Conclusion

  • AgroAskAI combines domain-specific tools, generative AI, specialized agents, and governance mechanisms to provide detailed, step-by-step climate-adaptation guidance.The conclusion reports superior actionable and localized recommendations compared with static modeling tools and generic single-agent AI models.Source: §5 · p. 7

Limitations

  • System performance is bounded by the quality and scope of external data, multilingual support depends on OpenAI’s language model, and low-connectivity deployment remains an operational challenge.The authors suggest centralized computing hubs as one possible shared-access approach for farmers without devices or internet connectivity.Source: §5 · p. 7

Future Work

  • Future work will test usability, accessibility, and cultural appropriateness with farmers through participatory design, field trials, and iterative refinement.Planned testing includes regions with varied literacy levels, dialects, and technological familiarity.Source: §5 · p. 7

A Pseudo Code for Agents in AgroAskAI

The pseudocode describes a workflow that parses farmer queries, routes them among specialized agents, and returns weather summaries or reviewed solution plans. Clarification loops and review approval are built into the described procedures.

PromptAgent and ParseAgent

  • The Prompt Agent detects the user’s language and either sends a query to parsing or asks for clarification; the Parse Agent extracts intent, location, and time, checks scope, and requests missing required details.After parsing, it passes the query and structured data to the Agent Manager.Source: A Pseudo Code for Agents in AgroAskAI · p. 9

AgentManager

  • The Agent Manager selects agents, maintains conversation history, applies termination logic, and invokes the Reviewer Agent when selection returns no next agent.The pseudocode lists Prompt, Parse, Solution, Weather History, Weather Forecast, and Reviewer agents.Source: A Pseudo Code for Agents in AgroAskAI · p. 9

Weather Agents

  • The Weather Forecast and Weather History agents use the parsed location and time to retrieve forecast or NASA data, then generate and return summaries.The forecast procedure uses forecast data; the history procedure uses NASA data.Source: A Pseudo Code for Agents in AgroAskAI · p. 9

SolutionAgent and ReviewerAgent

  • The Solution Agent generates a plan from the parsed query, adaptations, and available weather data, then sends it for review and waits for Agent Manager approval.The Reviewer Agent evaluates the plan against forecast, NASA, adaptation, and query information, returning either improvement instructions or an approval message.Source: A Pseudo Code for Agents in AgroAskAI · p. 10

B Sample Decision-Support by AgroAskAI

In the Guatemala pest-control example, the first response offered a broad IPM list and was reviewed for not answering the request for one method. A revised answer focused on intercropping and received reviewer approval.

Response 1

  • The initial response listed crop diversification, planting timing, biological controls, intercropping, sanitation, resistant varieties, and soil practices as components of IPM.It presented the techniques as a combined set of practical pest-control methods for Guatemala.Source: B Sample Decision-Support by AgroAskAI · p. 10

Review of Response 1

  • The reviewer found that the first IPM response offered multiple techniques without prioritizing one, omitted specific Guatemalan crops, and did not address farmers’ economic context.The review characterized the response as broad and insufficiently prioritized.Source: B Sample Decision-Support by AgroAskAI · pp. 10–11

Why Intercropping is Recommended

  • The revised answer described companion crops as deterring pests, crop mixing as disrupting pest life cycles, and intercropping as making efficient use of limited farm space.It also associated intercropping with diversified outputs, reduced pesticide use, and maize-bean combinations relevant to Guatemala.Source: B Sample Decision-Support by AgroAskAI · p. 11

Revised Response

  • The revised response selected intercropping as the single method requested and the Reviewer Agent approved the response.It connected the recommendation to Guatemala and named the Three Sisters strategy as a specific application.Source: B Sample Decision-Support by AgroAskAI · p. 11

C Comparison of CROPWAT, ChatGPT, and AgoAskAI

The comparison describes CROPWAT’s Kitui response as covering rainfall metrics and irrigation scheduling, while noting several kinds of advice it omits. The response examples also present recommendations on climate-resilient crops, water conservation, agroforestry, soil conservation, and farming practices.

C Comparison of CROPWAT, ChatGPT, and AgoAskAI

1. Adopt Climate-Resilient Crops

  • Drought-tolerant and early-maturing crops listed for farmers include sorghum, millet, pigeon peas, and improved maize varieties.Source: §1 · p. 12

2. Use Water Conservation Techniques

  • Water-conservation measures include zai pits or planting basins, mulching, and roof-water harvesting with storage tanks.Source: §2 · p. 12

3. Agroforestry

  • The agroforestry advice proposes Moringa, Neem, and Melia volkensii for soil fertility, shade, biodiversity, and windbreaks.Source: §3 · p. 12

4. Soil Conservation

  • Soil-conservation suggestions include terracing, cover cropping, composting, and minimum tillage.Source: §4 · p. 12

5. Improved Farming Practices

  • Improved farming practices include intercropping and crop rotation to reduce pest pressure, plus using seasonal forecasts to guide planting.Source: §5 · p. 12

6. Insurance & Risk Management

This section combines Kitui weather observations with advice on crops, planting, water, soils, livestock, and diversification. It also identifies insurance and farmer-support programs as risk-management resources.

6. Insurance & Risk Management

  • Risk-management resources include index-based crop insurance and mobile weather-advice platforms; the section also lists KCSAP training, NDMA early warnings, and FAO and World Bank programs.Source: §6 · p. 12
  • Kitui’s analyzed monthly rainfall ranged approximately from 0.08 mm to 7.33 mm during 2010–2022, while average temperatures ranged from 21°C to 27°C.Source: §6 · p. 12

1. Adopt Drought-Resistant Crops

  • Recommended drought-resistant crops include sorghum, millet, and cowpeas, alongside hybrid or drought-resistant seeds and crop diversification.Source: §1 · p. 12

2. Plant Timing and Early Maturing Crops

  • The planting advice is to sow shortly before expected rains and choose early-maturing crops to complete growth during rainy periods.Source: §2 · p. 12

3. Irrigation and Water Harvesting

  • Water strategies include small-scale drip or bucket irrigation, rainwater storage, rock catchments, and sand dams.Source: §3 · p. 12

4. Soil Health Management and Conservation

  • Soil management combines manure, compost, or fertilizers with terracing, mulching, and cover crops to support water retention and limit erosion.Source: §4 · p. 12

5. Switching or Supplementing with Livestock Keeping

  • The advice suggests supplementing crop farming with drought-tolerant goats or sheep suited to Kitui’s arid conditions.Source: §5 · p. 12

6. Mixed Cropping and Crop Diversification

  • Mixed cereals and legumes, along with high-value crops or vegetables, are proposed to diversify production and reduce complete-crop-failure risk.Source: §6 · p. 12

7. Community and Cooperative Participation

The section recommends farmer groups and extension support for timely updates and technical assistance. It also points to permaculture and agroforestry as longer-term resilience strategies.

7. Community and Cooperative Participation

  • Farmers are advised to join local weather-monitoring and farmer groups and engage extension officers for timely updates, resource sharing, and technical support.Source: §7 · pp. 12–13

8. Switch to Water-Efficient Practices in Farming

  • For a long-term strategy, the section names permaculture or agroforestry as practices that can improve resilience.Source: §8 · p. 13

Related papers

Loading 2512.14910v1…