JSAI Journal Repository

Abstract

Web application security remains a critical challenge in software development. Manual security audits are time-consuming and costly, while traditional Static Application Security Testing (SAST) tools often yield false positives and negatives. The code processing capabilities of Large Language Models (LLMs) offer a potential solution, but their effectiveness is highly dependent on the quality of the prompts provided. This experimental study applied a set of structured prompt engineering techniques to analyze security vulnerabilities in open-source code samples containing known vulnerabilities. The LLM model used was GPT-4. The procedure consisted of three stages: (1) Context Initialization, (2) Multi-Layer Analysis (general code review, specific vulnerability pattern recognition, and dependency check), and (3) Remediation Recommendation Generation. The designed prompt engineering techniques successfully identified the majority of known vulnerabilities, including Cross-Site Scripting (XSS), SQL Injection (SQLi), and Insecure Deserialization. The model also generated specific and implementable code repair recommendations, such as the use of prepared statements and output sanitization functions. The results indicate that a structured prompt engineering approach can be an effective assistive tool for initial security audits, capable of reducing the analytical burden on developers. However, this approach has limitations, including the potential for false positives, an absolute dependence on prompt quality, and the model's tendency to hallucinate on highly complex code. This study underscores the need for a standardized prompt framework for cybersecurity.

Keywords

Abstract

This study aims to analyze and formulate effective prompt structures for the DeepSeek Large Language Model (LLM) to enhance efficiency and quality in web application development.
This research employs a Research and Development (R&D) approach with an experimental method. A series of common web development tasks (code generation, debugging, documentation, and architecture) were tested on DeepSeek. Each task was executed using varied prompts, including zero-shot, few-shot, and chain-of-thought prompting. Effectiveness was measured based on the functional accuracy of the code, completeness of output, and the number of iterations required for task completion. Structured and contextual prompts produced significantly superior outputs. The chain-of-thought technique proved most effective for complex tasks like architecture design and debugging, improving logic accuracy by up to 40% compared to zero-shot. Meanwhile, few-shot prompting was optimal for generating code with consistent style and standards. Providing clear project context (technology stack, API specifications, and constraints) reduced iteration needs by up to 60%. Utilizing DeepSeek Gen.AI for web development requires deliberate prompt strategies. A prompt formulation that includes the AI's role, detailed context, the expected output format, and critical thinking steps is a determining factor for success. These findings provide a practical framework for developers to leverage LLMs more productively.

Keywords

Abstract

Penerimaan Mahasiswa Baru (PMB) merupakan proses rutin tahunan yang vital di setiap perguruan tinggi. Di Politeknik Negeri Padang (PNP), salah satu jalur masuk adalah Jalur Seleksi Mandiri (SMPNP). Selama ini, proses pendaftaran yang manual dinilai kurang efektif, menyebabkan calon mahasiswa harus datang langsung ke kampus dan berpotensi menimbulkan antrian panjang. Penelitian ini bertujuan untuk merancang dan membangun aplikasi pendaftaran online guna mengatasi permasalahan tersebut. Metode penelitian yang digunakan meliputi studi literatur, analisis kebutuhan, perancangan sistem, dan implementasi. Aplikasi dirancang menggunakan CMS Joomla dengan komponen FacileForms untuk pembuatan formulir, PHP sebagai bahasa pemrograman, dan MySQL untuk manajemen basis data, yang berjalan pada sistem operasi server Linux CentOS. Aplikasi ini mengintegrasikan proses permintaan PIN, pembayaran biaya pendaftaran secara host-to-host dengan Bank Nagari Sumatera Barat, pengisian borang data diri dan pilihan program studi, hingga pencetakan Kartu Tanda Pendaftaran. Hasil dari penelitian ini adalah sebuah aplikasi web yang berfungsi penuh dan telah diimplementasikan pada https://pmb.pnp.ac.id. Aplikasi ini telah berhasil digunakan pada Penerimaan Mahasiswa Baru Tahun Akademik 2023/2024, yang memberikan kemudahan bagi calon mahasiswa, efisiensi bagi bagian akademik, dan akurasi pelaporan keuangan. Ke depan, pengembangan sistem berbasis mobile dan integrasi SMS Gateway untuk pengumuman kelulusan sangat disarankan.

Keywords

Abstract

The development of web-based scientific journal repository systems is an urgent need in supporting knowledge dissemination in the digital era. This study aims to develop a scientific journal repository application with three-tier architecture and comprehensive role-based access control (RBAC) implementation. The development method uses a Research and Development (R&D) approach with a modified waterfall model. The system was developed using native PHP, MySQL, and Bootstrap 5 technologies by implementing cybersecurity principles and optimal user experience. The research results produced a repository application consisting of: (1) responsive frontend for journal publication; (2) backend with three different roles (admin, author, reviewer); (3) integrated review system; (4) secure user management; and (5) efficient search and pagination features. Testing showed the system successfully handled all use cases with 98.7% success rate, average response time of 1.2 seconds, and zero critical vulnerabilities. This application has been implemented in production at https://jsai.parewa.my.id and proven effective in supporting digital scientific journal publication processes.

Keywords

Abstract

Developing vocational education curricula responsive to Industry 4.0 dynamics is an urgent need. This study aims to design an Outcome-Based Education (OBE)-based curriculum for the D3-Computer Engineering Study Program that integrates the Indonesian National Qualification Framework (KKNI) Level 5, digital industry needs, and institutional vision. The research method uses Research and Development (R&D) with a modified ADDIE model. Data was collected through documentation studies, curriculum benchmarking, FGDs with industry stakeholders, and design simulation assisted by Generative AI DeepSeek. The results produced a comprehensive OBE curriculum model including: (1) strengthened vision-mission formulation; (2) five Program Educational Objectives (PEO); (3) five Graduate Profiles; (4) twelve Graduate Learning Outcomes (GLO); (5) forty Course Materials; and (6) an operational curriculum structure of 110 credits with 53.6% practical composition. The curriculum integrates Industry 4.0 competencies such as Internet of Things (IoT), embedded systems, cybersecurity, cloud computing, and technopreneurship. Correlation matrix analysis shows high consistency between components with an average TPP-GP correlation coefficient of 0.82. The use of Gen AI proved effective in accelerating the analysis process by 60% and ensuring consistency of vertical integration. This curriculum model is recommended for implementation in computer engineering vocational study programs with institutional context adjustments.

Keywords