Translating agricultural Packages of Practices (PoPs) into local languages is crucial for knowledge dissemination but poses challenges in terms of cost, time, and accuracy, particularly for domain-specific terminology. This paper investigates automating PoP translation using Large Language Models (LLMs), specifically Gemini-1.5-Flash. Initial experiments revealed inaccuracies in translating agricultural terms. To address this, we propose a novel dictionary-based approach, integrating a specialised agricultural dictionary (approx. 10,000 terms) with the LLM. The method employs embeddings (“all-MiniLM-L6-v2”) and K-Means clustering to efficiently retrieve and provide contextually relevant dictionary terms to the LLM during translation. This significantly reduces computational overhead compared to a linear search. Results demonstrate a substantial improvement in translation accuracy for Sinhala and Tamil, validated by domain experts. This optimised dictionary-enhanced LLM approach offers an effective solution for accurate and efficient automated translation of specialised agricultural content.