Independent systems builder
Portfolio / 2026
Indonesia / UTC+7
Deep Learning SystemsArchitecture
I build tools for the work around deep-learning models: testing them, running them, tracing failures, and giving people a clear way to control them.
Current direction
Local, observable AI tooling.
A prompt goes in. An answer comes out. The demo works. The real engineering has not started.
What changed? Why did it fail? What did it cost? Can a person steer it?
I work on evaluation, runtime behavior, memory, tooling, and the interface that lets a person inspect and steer the system.
About / 02
How I make decisions.
Indonesia / UTC+7
JinXSuper / Jin
Independent systems builder
I stay with a problem after the demo: through edge cases, instrumentation, evaluation, and the moment someone has to decide whether the result deserves trust.
I prefer explicit trade-offs, quiet tools, and software that tells you what it is doing. Complexity does not need to disappear; it needs to be inspectable.
Working biasEvidence over spectacle. Explicit over magical.
Selected work / 03
My partSystem architecture, product direction, Rust implementation, and interaction model.
Rust · AI tooling · Systems design
My partProduct architecture, desktop experience, accessibility model, and TypeScript implementation.
TypeScript · Desktop · Accessibility
My partArchitecture, native engine, memory model, and product implementation.
Rust · Svelte · Agent systems
Tools / 04
Current stack / languages + ways of seeing the system
The engine, interface, and evidence need different tools. Rust handles native work; TypeScript and Svelte shape the product surface; evaluation, local memory, and observability keep the system understandable.
Hover, focus, or tap a tool to shift the focal plane.
01 / Native engine
For native engines, local runtimes, and the code paths where predictability and performance matter.
Contact / 05
Available for systems work.
Indonesia / UTC+7
A benchmark that no longer reflects the product. A runtime with no useful trace. An interface that asks for trust before it earns it. Those are good places to start.
Send the messy version.
Email works best.