<![CDATA[Cocoa (Theobroma cacao L.) represents a strategic agricultural commodity supporting rural smallholder livelihoods across Bali Province, notably in Tabanan, Jembrana, Gianyar, and Buleleng regencies. Nevertheless, farm productivity and bean quality remain severely threatened by Cocoa Pod Borer (Conopomorpha cramerella), Helopeltis sp., and black pod rot (Phytophthora palmivora), exacerbated by the limited availability of agricultural extension officers in remote rural areas. This study develops Kakao Assistant, a cross-platform mobile virtual assistant built on Flutter integrating Google Gemini 3.1 Flash Large Language Model (LLM) under a single-role architecture dedicated to Cocoa Farmers. System development follows the Modified Waterfall lifecycle comprising requirements analysis, system architecture design, implementation, and empirical testing. The application provides multimodal agronomic consultations (text inquiries and pod disease image diagnostics), regional agroclimatic contextualization for Bali plantations, Text-to-Speech (TTS) voice synthesis, message bookmarking, and an on-device model evaluation module ensuring transparency of artificial intelligence inferences. Black-box testing validated that 100% of the 11 functional scenarios functioned in strict compliance with system specifications. Prototype testing verified the module's capability to track response latency, network fault resilience, and local FIFO history logs prior to extensive community deployment across Bali cocoa farms.]]>
Copyrights © 2026
