Vyncint Ng

Rust systems developer · GPU tooling · terminal testing infrastructure · Kubernetes · Go

Creator of termlens · #2 contributor to NVlabs/cuda-oxide · #7 contributor to kube-logging/logging-operator

GitHub followers crates.io Việt Nam
Rust CUDA PTX Metal Apache Arrow Kubernetes Go Swift

I build Rust tools for testing terminal applications and analysing GPU kernels, and I work on Kubernetes operators and sandboxed agent runtimes. Most of my contributions are testing infrastructure and CI: making a tool's claims checkable, and catching the places where its documentation and its code disagree.

Projects

Six Rust products on crates.io. termlens tests the other five end to end: it runs each real binary in a pseudo-terminal and asserts on the rendered screen.

Testing

termlens

Headless end-to-end testing for CLI and TUI apps: real PTYs, rendered-screen assertions, snapshots, screen diffs. The end-to-end example in Ratatui's snapshot-testing recipe.

version downloads CI docs
GPU · static analysis

reconverge

Static analysis for Rust GPU kernels: divergent barriers and warp-collective hazards, explained lane by lane. No GPU required.

version downloads CI
GPU · autotuning

launchbound

Autotuning for GPU kernels: a convergence check per configuration, CUDA and Metal benchmarking, analytical estimates.

version downloads CI
Deep learning

oxmera

Tensor and deep-learning framework: autograd, CPU, Metal and CUDA backends, a terminal training dashboard.

version downloads CI
Data

oxidelake

Arrow-native SQL engine on DataFusion or Ballista, with optional GPU acceleration and per-batch CPU fallback.

version downloads CI
Terminal

mossaic

Plan, draw and track GitHub contribution art, with pixel-rendered graphs in the terminal.

version downloads CI
Go · sandboxes

openshell-driver-applecontainer

A compute driver for NVIDIA OpenShell backed by apple/container: one micro-VM per sandbox on Apple silicon, roughly one second to Ready.

stars

Upstream

How I work

Evidence over assertion

A bug report states what was measured, against which released version.

A check must be seen to fail

Every new CI check is broken on purpose once, to prove it can fail, before it is trusted.

Say what you did not do

A pull request lists what it left out and why. A known gap is cheap; a false claim is expensive.

AI-assisted, human-owned

I use AI tools in development and take responsibility for reviewing, testing and maintaining the results.

More on how these projects are built

Contact

Open to collaboration on Rust tooling, GPU infrastructure, Kubernetes operators and testing. Open an issue or discussion on any repository above, or find me at github.com/vyncint.