Rancang Bangun Backend dan Chatbot untuk Aplikasi Taksasi Produksi Kelapa Sawit

Hakim, Muhammad Irfan (2026) Rancang Bangun Backend dan Chatbot untuk Aplikasi Taksasi Produksi Kelapa Sawit. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221291-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Perkebunan kelapa sawit memiliki peran penting dalam perekonomian Indonesia, sehingga proses taksasi produksi yang akurat dan efisien diperlukan untuk mendukung perencanaan operasional serta pengambilan keputusan. Selain itu, kebutuhan terhadap akses informasi yang cepat dan kontekstual terkait kondisi perkebunan juga diperlukan, sehingga sistem yang tidak hanya mampu melakukan estimasi produksi, tetapi juga membantu pengguna memperoleh informasi pendukung dan rekomendasi secara lebih mudah. Namun, metode taksasi konvensional masih menghadapi kendala berupa proses yang memerlukan waktu lama serta rentan terhadap kesalahan manusia. Penelitian ini bertujuan merancang dan mengembangkan sistem taksasi produksi kelapa sawit berbasis backend dengan arsitektur hibrida yang menggabungkan pendekatan modular monolithic dan microservices. Sistem yang dikembangkan memisahkan layanan utama yang memiliki keterkaitan erat dengan kebutuhan bisnis ke dalam arsitektur modular monolithic, sedangkan layanan dengan kebutuhan komputasi khusus seperti tree counting, prediksi produktivitas, rekonstruksi tiga dimensi, dan chatbot diimplementasikan sebagai microservices terpisah. Chatbot dikembangkan sebagai asisten cerdas berbasis Agentic Retrieval-Augmented Generation (Agentic RAG) yang membantu pengguna memperoleh informasi umum, informasi analitik internal, serta rekomendasi berdasarkan konteks perkebunan. Hasil implementasi menunjukkan bahwa seluruh layanan berhasil terintegrasi dan berjalan sesuai kebutuhan fungsional. Berdasarkan hasil load testing dan stress testing, sistem mampu mempertahankan waktu respons rata-rata di bawah 650 ms pada kondisi beban normal dengan tingkat kesalahan maksimum sebesar 0,0322%, serta tetap mempertahankan tingkat kesalahan di bawah 0,1% ketika terjadi peningkatan beban sebesar 80,04%. Evaluasi menggunakan RAGAS, BERTScore, dan human evaluation menunjukkan bahwa chatbot mampu menghasilkan respons yang relevan, konsisten terhadap konteks, dan menghasilkan query basis data internal yang valid melalui mekanisme text-to-SQL. Dari seluruh skenario pengujian chatbot, model Qwen3:14B dipilih sebagai model terbaik karena mampu menghasilkan respons dengan Tingkat akurasi, relevansi, dan kesesuaian konteks yang paling optimal.
========================================================================================================================================
Oil palm plantations play an important role in Indonesia’s economy, making accurate and efficient production estimation processes necessary to support operational planning and decision-making. In addition, the need for fast and contextual access to plantation-related information has increased, requiring a system that not only provides production estimation but also assists users in obtaining supporting information and recommendations more effectively. However, conventional production estimation methods still face challenges due to time-consuming processes and susceptibility to human errors. This study aims to design and develop a backend-based oil palm production estimation system using a hybrid architecture that combines modular monolithic and microservices approaches. The developed system separates core services with strong business dependencies into a modular monolithic architecture, while services with specific computational requirements such as tree counting, productivity prediction, three-dimensional reconstruction, and chatbot are implemented as independent microservices. The chatbot is developed as an intelligent assistant based on Agentic Retrieval-Augmented Generation (Agentic RAG) to assist users in obtaining general information, internal analytical information, and context-based recommendations related to plantation conditions. The implementation results show that all services are successfully integrated and operate according to the defined functional requirements. Based on load testing and stress testing results, the system maintains an average response time below 650 ms under normal workload conditions with a maximum error rate of 0.0322%, while maintaining an error rate below 0.1% under an 80.04% increase in workload. Evaluation using RAGAS, BERTScore, and human evaluation demonstrates that the chatbot generates relevant and context-consistent responses and successfully produces valid internal database queries through the text-to-SQL mechanism. Among all chatbot evaluation scenarios, the Qwen3:14B model is selected as the best-performing model because it produces responses with optimal accuracy, relevance, and contextual alignment.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Agentic RAG, chatbot, kelapa sawit, layanan mikro, modular monolith, taksasi produksi, Agentic RAG, chatbot, microservices, modular monolith, oil palm, production estimation
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > T Technology (General) > T58.6 Management information systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Irfan Hakim
Date Deposited:	27 Jul 2026 04:13
Last Modified:	27 Jul 2026 04:13
URI:	http://repository.its.ac.id/id/eprint/137505

Actions (login required)

View Item