Computer Science undergraduate from Indonesia. I work on computer vision, NLP, and edge AI, mostly getting models to run on small hardware, so they keep working without a network.
I also shoot photos, work on audio, and build mobile apps, and a lot of my projects sit between those and ML.
Currently exploring on-device LLM efficiency and multimodal edge AI. Open to internships and collaborations in Edge AI / MLOps, 📫 .
|
Context-aware interactive robot assistant split across a laptop and a Raspberry Pi 4: CUDA inference on the laptop, the Pi driving the camera, audio I/O, and hardware control. Voice, vision, and gesture are fused into one state, so it responds to what it sees and hears together rather than to whichever sensor fires first.
|
RAG chatbot for media-framing analysis of Indonesian English-language news. Runs fully local: a 3B LLM on Ollama, and embeddings stored in PostgreSQL as float arrays, with no external API at inference. Every answer is scored twice, trained model against a rule-based baseline, side by side. That comparison is the point: it keeps the cost of the simple heuristic visible instead of assuming the trained model wins.
|
|
Pothole volumetric analysis for road maintenance. Combines monocular depth estimation with instance segmentation to measure pothole volume from a single camera.
|
Local translation engine on Meta's NLLB-200, which covers 200 languages, running entirely on your own hardware. No data leaves the machine.
|
|
Real-time multi-modal facial analysis from a webcam: emotion, drowsiness, and a pain proxy. The first published numbers were wrong and I said so:
|
Keyboard-first desktop app for culling large photo shoots fast. Built for my own photography workflow, where the bottleneck is the first pass, not the editing.
|
