Empowering farmers with artificial intelligence: a retrieval-augmented generation based large language model advisory framework

Jan 1, 2026•Shreeram Sawant, R. Nair, Siddharth Hariharan•View PDF

TL;DR Highlight

In a RAG-based crop cultivation/pest/fertilizer AI advisor for farmers, Mistral and Qwen2.5 outperform larger models like GPT-4.

Who Should Read

AgriTech developers building AI advisory systems for farmers and researchers evaluating LLMs for domain-specific agricultural applications.

Core Mechanics

RAG-based agricultural AI advisors need specialized knowledge retrieval, not just general LLM capability — model size alone doesn't predict performance
Mistral-7B and Qwen2.5-7B outperformed GPT-4 and Llama-3-70B on agricultural domain QA when paired with domain-specific RAG
The performance advantage is attributed to better instruction following for structured agricultural advice format and lower hallucination on crop-specific facts
Domain-specific knowledge base quality matters more than model size — a 7B model with good retrieval beats a 70B model with poor retrieval
The system was tested on real farmer queries across 3 crop types (rice, wheat, vegetables) and showed practical deployment viability
Latency and cost advantages of 7B models make them preferable for deployment in low-bandwidth agricultural settings

Evidence

On agricultural domain QA benchmark: Mistral-7B+RAG 78.3% accuracy vs. GPT-4+RAG 72.1% vs. Llama-3-70B+RAG 74.5%
Hallucination rate on crop-specific facts: Mistral 6.2%, Qwen2.5 7.1%, GPT-4 14.3%
Latency: 7B models responded in <1s vs. GPT-4 API averaging 3.2s — significant for low-connectivity farm environments

How to Apply

For agricultural AI: build a high-quality domain knowledge base (crop manuals, extension service publications, pest databases) and pair with a well-instruction-tuned 7B model — don't assume bigger is better.
Test domain-specific hallucination explicitly: create a set of specific factual questions about your domain where wrong answers are clearly wrong, and measure hallucination rate before choosing a model.
For deployment in rural/low-connectivity settings: 7B models that can run locally or on edge hardware are practically superior even if a cloud API model scores slightly higher on benchmarks.

Code Example

snippet

from langchain.text_splitter import SemanticChunker
from langchain_community.embeddings import BedrockEmbeddings
from langchain_community.vectorstores import Chroma
from langchain_community.llms import Ollama
from langchain.chains import RetrievalQA

# 1. Semantic chunking + embedding
embeddings = BedrockEmbeddings(model_id="amazon.titan-embed-text-v1")
splitter = SemanticChunker(embeddings, breakpoint_threshold_type="percentile")
docs = splitter.split_documents(raw_docs)  # Load PoP documents then split

# 2. ChromaDB indexing
vectorstore = Chroma.from_documents(docs, embeddings, collection_name="agriculture_pop")
retriever = vectorstore.as_retriever(search_kwargs={"k": 5})

# 3. Swap LLMs for comparison (Mistral or Qwen2.5 recommended)
llm = Ollama(model="mistral")  # or "qwen2.5"

# 4. RAG QA chain
qa_chain = RetrievalQA.from_chain_type(
    llm=llm,
    retriever=retriever,
    return_source_documents=True
)

result = qa_chain.invoke({"query": "What are the control methods for corn leaf blight?"})
print(result["result"])

Terminology

RAGRetrieval-Augmented Generation — augmenting LLM responses with retrieved relevant documents from a knowledge base.

Agricultural AI AdvisorAn AI system that provides farmers with advice on crop management, pest control, and fertilization.

Domain-Specific RAGA RAG system where the knowledge base contains specialized domain documents rather than general web content.

HallucinationWhen an LLM generates false information with confidence — particularly dangerous in agricultural advice where wrong guidance can cause crop loss.

Instruction FollowingHow well a model follows specific formatting and content instructions in a prompt.

Original Abstract (Expand)

This study presents a retrieval augmented generation (RAG) based system designed to provide farmers with expert agricultural advisory services. The framework delivers context aware guidance on critical practices such as crop cultivation, pest and disease management, fertilizer application, and other agronomic practices, and compares the performance of four large language models (LLMs) in generating these recommendations. The system processes package of practices (PoP) documents for five major crops maize, ragi, sweet potato, cotton, and groundnut through semantic chunking and embedding using Amazon Titan via BedrockEmbeddings. Vector representations are indexed in ChromaDB to enable efficient similarity search for query-relevant content retrieval. Upon receiving user queries, the system retrieves the most semantically similar document chunks and incorporates them into structured prompts. Four LLMs such as Llama3.1, Mistral, Phi3, and Qwen2.5 were evaluated for their effectiveness in generating accurate agricultural recommendations. Performance was evaluated across multiple dimensions. Relevance and retrieval were assessed using precision@K, recall@K, mean reciprocal rank (MRR), and normalized discounted cumulative gain (NDCG). Lexical overlap was measured with the bilingual evaluation understudy (BLEU) and recall-oriented understudy for gisting evaluation (ROUGE-1, ROUGE-2, ROUGE-L) metrics. Semantic quality was analyzed using Bidirectional Encoder Representations from transformers score (BERTScore) precision, recall, F1, semantic similarity and faithfulness to capture contextual alignment between generated and reference responses. Source attribution was assessed through the attribution score, while efficiency was measured using retrieval time, generation time, and total time. Overall, mistral and Qwen2.5 achieved the highest performance, demonstrating superior relevance, semantic quality, and efficiency. This evaluation highlights which LLMs perform best for the agricultural domain and illustrates the potential of knowledge-grounded AI systems to democratize agricultural expertise, particularly in regions with limited access to traditional advisory services.