Wasserstein GAN and Conditional WGAN-GP trained to synthesise high-quality data for profession prediction across 8 imbalanced classes.
- Python
- PyTorch
- TensorFlow
- GAN
- WGAN
- CWGAN-GP
Depok, Indonesia
AI/ML engineer, backend engineer, full-stack developer, and ML researcher.
I build AI products, backend services, and the web platforms that sit on top of them — from the research notebook to the thing people actually use.
what I do
I build AI products end to end: the model, the service behind it, and the interface in front of it. Most of my work lives in Python and Go, across the whole machine-learning cycle — GAN-based synthetic data, computer vision, RAG and multi-agent systems. Right now that means co-founding Decklify, an AI SaaS that turns prompts, documents and web research into presentations. The parts I enjoy most are the ones that do not photograph well: the training pipeline that has to survive a retrain, the API that stays up when the model behind it does not.
selected projects
3 featured projects selected projects
Wasserstein GAN and Conditional WGAN-GP trained to synthesise high-quality data for profession prediction across 8 imbalanced classes.
AI SaaS that turns prompts, documents and web research into sourced, narration-ready slides — ingest, RAG search, slide pipeline, in-browser editor.
End-to-end manga translator pairing Japanese OCR with automatic translation, plus a Kaggle pipeline for fine-tuning the OCR model.
Case studies, source and project links
the tools
the long version
The long version. Same person, more rows — the full record as a document you can print or download.
get in touch
Got something that needs building, breaking, or explaining badly? The window is open.