Laporan kerja Praktek Sistem Rekomendasi Pola Servis Sepeda Motor Otomatis Berbasis Catboost

Nicholas, Gerry (2026) Laporan kerja Praktek Sistem Rekomendasi Pola Servis Sepeda Motor Otomatis Berbasis Catboost. Project Report. [s.n.]. (Unpublished)

Text 5025231017-Project_Report.pdf Restricted to Repository staff only Download (978kB)

Abstract

Identifikasi jadwal servis berikutnya dan rekomendasi penggantian part pada sepeda motor selama ini banyak bergantung pada estimasi manual mekanik/service advisor, tanpa memanfaatkan riwayat servis pelanggan secara sistematis, sehingga rekomendasi antar bengkel resmi kurang konsisten. Untuk mengatasinya, disusun sistem rekomendasi pola servis berbasis model machine learning CatBoost yang dilengkapi lapisan large language model (LLM) untuk menghasilkan narasi rekomendasi berbahasa natural. Dokumen desain awal sistem ini masih berupa arsitektur satu server FastAPI tanpa mempertimbangkan aspek operasional produksi. Kerja praktik ini berfokus pada perancangan dan implementasi lapisan infrastruktur yang mengoperasionalkan rancangan tersebut, mencakup containerization dengan Docker, reverse proxy dan load balancing dengan Traefik, penskalaan horizontal beberapa replika backend, integrasi awal LLM (klien OpenAI-compatible) untuk narasi rekomendasi beserta mekanisme fallback-nya, serta pengujian beban (load testing) dan riset autoscaling menggunakan k6. Hasil pengujian beban menunjukkan jalur inferensi cepat (/predict?fast=true) tetap stabil hingga 16 virtual user (RPS ±38, p95 499,51 ms) sebelum mengalami kegagalan pada beban 32 virtual user, sementara pembuktian latency melalui logging waktu menunjukkan bahwa mayoritas waktu tunggu pengguna (>5 detik) berasal dari pemanggilan API eksternal milik klien, bukan dari komputasi inferensi model CatBoost yang berjalan pada skala milidetik. Hasil kerja praktik ini menghasilkan fondasi infrastruktur deployment yang siap dijalankan pada Virtual Private Server (VPS) dan dapat diskalakan sesuai kebutuhan trafik bengkel resmi.
===================================================================================================================================
Identification of the next service schedule and part replacement recommendations for motorcycles has largely relied on manual estimation by mechanics/service advisors, without systematically leveraging customer service history, resulting in inconsistent recommendations across authorized service centers. To address this, a service pattern recommendation system was developed based on a CatBoost machine learning model equipped with a large language model (LLM) layer to generate natural-language recommendation narratives. The system's initial design document consisted only of a single-server FastAPI architecture without consideration for production operational aspects. This internship (Kerja Praktik) focused on the design and implementation of the infrastructure layer that operationalizes that design, covering containerization with Docker, reverse proxy and load balancing with Traefik, horizontal scaling of multiple backend replicas, initial LLM integration (OpenAI-compatible client) for recommendation narratives along with its fallback mechanism, and load testing and autoscaling research using k6. Load testing results show that the fast inference path (/predict?fast=true) remained stable up to 16 virtual users (RPS ±38, p95 499.51 ms) before failing at a load of 32 virtual users, while latency evidence from time logging shows that the majority of user wait time (>5 seconds) came from calls to the client's external API, not from CatBoost model inference computation, which runs at millisecond scale. The results of this internship produced a deployment infrastructure foundation ready to run on a Virtual Private Server (VPS) and scalable according to authorized service center traffic needs.

Item Type:	Monograph (Project Report)
Uncontrolled Keywords:	Docker, Traefik, load balancing, deployment, large language model, CatBoost
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Gerry Nicholas
Date Deposited:	31 Aug 2026 05:30
Last Modified:	31 Aug 2026 05:30
URI:	http://repository.its.ac.id/id/eprint/144421

Available Versions of this Item

• Laporan kerja Praktek Sistem Rekomendasi Pola Servis Sepeda Motor Otomatis Berbasis Catboost. (deposited 28 Aug 2026 07:39)
  • Laporan kerja Praktek Sistem Rekomendasi Pola Servis Sepeda Motor Otomatis Berbasis Catboost. (deposited 31 Aug 2026 05:30) [Currently Displayed]

Actions (login required)

View Item