This project focuses on building a complete end-to-end machine learning pipeline with an emphasis on reproducibility, experimentation, and ML lifecycle best practices. The pipeline covers data versioning, model training, evaluation, and experiment tracking, enabling structured and repeatable machine learning workflows.

The system is designed using a DAG-based orchestration approach to manage dependencies between pipeline stages, ensuring reliability and scalability. This project demonstrates practical MLOps fundamentals and production-oriented machine learning development.

Repository: https://www.twine.net/signin

This project focuses on developing a multimodal deepfake detection system by combining visual and audio-based deep learning models. The approach integrates video features extracted using TimeSformer and audio features extracted using ECAPA-TDNN, enabling robust detection of manipulated media content.

The system leverages cross-modal feature fusion to capture inconsistencies between audio and visual signals, improving detection accuracy compared to single-modality approaches. This project demonstrates the application of advanced deep learning architectures for real-world AI security and media forensics use cases.

Repository: https://www.twine.net/signin

This project focuses on building an end-to-end sentiment analysis system for Indonesian e-commerce reviews collected from Tokopedia. The pipeline covers data collection, text preprocessing, feature extraction, and multi-class sentiment classification using machine learning and natural language processing techniques.

The objective of this project is to automatically identify customer sentiment (positive, neutral, negative) to support business insights, product evaluation, and customer experience analysis. The system is designed with reproducibility and scalability in mind, making it suitable for real-world applications and further deployment as an API-based service.

Repository: https://www.twine.net/signin

Developed an AI-powered pipeline for extracting structured knowledge graphs and analyzing personality traits (Big Five/OCEAN model) from natural language text.
The system integrates Large Language Models (LLMs) via Gemini and OpenAI APIs to enhance semantic understanding and text reasoning.
It includes graph-based visualization, automated personality profiling, and report generation modules, enabling interpretable and scalable analysis for research and educational use cases.

For Code in My Github:
https://www.twine.net/signin