/RIZ-kee/
Jakarta, Indonesia
English & Bahasa Indonesia
Rizky Azmi Swandy,SoftwareEngineer,AI/ML
NowNebula Data Servicessince April 2026
I cover the engineering side of Nebula’s global AI platform: contributing to it, maintaining what runs, and designing what’s next. Across Southeast Asia I also work on architecture and the client side.
Selected work
Karya pilihan
Applied-AI systems from platform, product, and research work. Fuller write-ups are on the projects page.
Making small language models keep up under load
A custom KV cache layer on the inference path of the in-house Toba models.
AI agents for a cultural library, from retrieval to bare metal
An agent workflow: retrieval over a digital library of Indonesian culture, payments, and self-hosted production hardware.
A knowledge corpus on Indonesian culture, built at scale
Distributed crawling feeding LLM-based article generation on Budaya Indonesia.
Real-time poultry monitoring on edge devices
From camera feeds to annotated data to YOLOv8 detection running on NVIDIA Jetson Nano.
Indonesian speech-to-speech on iOS
Apple's speech recognition, Gemini Pro, and Google text-to-speech chained into one voice pipeline.
Reading Batak script for a learning platform
An OCR pipeline for Batak script, from a research collaboration with Leiden University.
Writing
Tulisan
Crawlers, retrieval, small models, and KV cache bugs, in English and Bahasa Indonesia.
Retrieval tanpa embedding
The Smallest Model I’ve Ever Shipped
Understanding KV Cache Bugs in Transformers
Open source
Sumber terbuka
MacMultipeer Bundle (opens in new tab)
a Swift bridge that exposes encrypted Apple Multipeer Connectivity sessions to Unity on macOS.