skip to content
Liu Yang's Blog

Selected Work

Project

Graduate student focused on LLM agents, with previous work in full-stack web development, computer vision, and practical infrastructure tooling.

I enjoy building automation that improves everyday computing workflows, especially systems that make connectivity, operations, and personal infrastructure simpler to run.

GitHub Blog

Current Focus

  • / Developer tooling and workflow automation
  • / LLM agents and practical automation
  • / Computer networking, self-hosted services, and personal infrastructure
  • / End-to-end connectivity, stability, and bandwidth efficiency

Open Source

Public repositories that best represent infrastructure work, agent tooling, and applied ML experimentation.

View all repositories

Connectivity

openconnect-ssh

Lightweight SSH wrapper for AnyConnect and OpenConnect environments with stable recovery and network-aware forwarding.

  • + Automatic VPN recovery
  • + Split tunneling
  • + Reverse port forwarding
Shell SSH OpenConnect

Agent Tooling

codex-telegram-bridge

Self-hosted Telegram bridge for persistent Codex conversations, streaming replies, queueing, cancel, and steer controls.

  • + Persistent sessions
  • + Streaming responses
  • + Operator controls
Python Telegram Codex

Applied ML

AnnounceChat

Announcement-oriented QA and prompt-engineering experiments with preprocessing pipelines for practical dataset work.

  • + Dataset preprocessing
  • + Prompt experiments
  • + Jupyter workflows
Python Jupyter

Private / Closed-Source Work

Selected internal and private projects summarized in a public-safe way, with concrete scope, system design, and implementation focus.

Shareable summaries of non-public work

Network Systems

XLAN.NET

Traffic control and traffic-splitting framework built around TUN interfaces and policy routing, designed for scalable dual-stack optimization across mixed TCP and UDP workloads.

  • + Client and server rule engines based on protocol, process, CIDR, and domain
  • + Custom UDP-over-Anything transport experiments for weak-network stability and latency
  • + OpenWRT deployment with extensive documentation on routing, encapsulation, and network tuning
TUN Policy Routing TCP/UDP IPv4/IPv6 OpenWRT

DevOps

DevOps Pipeline

Automated deployment pipeline built with Gitea, Drone, and Docker to reduce repetitive testing and release steps across environments.

  • + Code push to image build to container deployment automation
  • + One-click rollout flow for multiple environments
  • + Lowered manual deployment and verification overhead
Gitea Drone Docker CI/CD

Developer Tooling

Gost Port Forwarding

Port-forwarding and secure environment-sharing utility based on Gost and SSH, used to expose shared test services without duplicating middleware stacks.

  • + Keepalive and auto-reconnect for stable port mapping
  • + SSH and UFW based secure access for centralized middleware exposure
  • + Lightweight CLI and startup scripts for cross-machine testing workflows
Gost SSH UFW Shell CLI

Full-Stack Product

Memory

Campus Training Project - First Prize

User-driven foreign language learning community built from scratch as a campus training project, combining content, social interaction, and real-time communication.

  • + Full-stack system built with Spring Boot and Vue
  • + Social, content, and P2P real-time communication via WebSocket and WebRTC
  • + Extensible vocabulary memory model plus end-to-end DevOps setup with DNS, CDN, HSTS, and monitoring
Spring Boot Vue WebSocket WebRTC DNS/CDN

Distributed Systems

OpenClaw Hydra

Distributed networking and compute scheduling architecture designed around OpenClaw, separating control, GPU compute, and IPv6 edge resources for flexible deployment.

  • + Three-server architecture for control, GPU compute, and IPv6 edge access
  • + OpenClaw integration with jump-host based access and OpenConnect route slicing
  • + IPv6-only nodes used as cost-effective compute capacity for CPU-intensive workloads
OpenClaw GPU IPv6 OpenConnect Jump Host

Working Style

I care about systems that are easy to operate, easy to recover, and useful in daily work. That usually means practical automation, stable infrastructure, and tooling that reduces manual overhead instead of adding more ceremony.

Open to collaborating on infrastructure tooling, web projects, and applied ML systems.

Toolbox

Python Java C Go SQL PyTorch Vue React Shell Linux Docker Git OpenWRT Cloud infrastructure