Zhejian Zheng
I build reliable agents, backend systems, and digital products between Sydney and Ningbo.
My work connects engineering depth with clear product decisions—from recoverable AI agents and event-driven systems to interfaces people can understand.
Sydney, NSW
Agent engineering · Backend systems · Product interfaces
Selected builds
Three projects that show how I reason about systems, safety, and useful AI.
Solana OrderFlow: Event-Driven Escrow with Anchor and Kafka
A Solana escrow project that connects an Anchor smart contract with Kafka consumers, Postgres projections, risk rules, and notifications.
Safe Reinforcement Learning with a Supervised Shield
A grid-world experiment where Q-learning is combined with a supervised safety classifier to reduce hazard violations.
GitHub Repo Review Agent: Engineering an AI Code Review Workflow
A bilingual engineering note on building an AI agent that reviews GitHub repositories through indexing, static analysis, verification, and risk-ranked findings.
Latest notes
Recent source-level writing for engineers choosing and building agent systems.
LLM Wiki in Practice: Turning Documents into Traceable Engineering Knowledge
How I use an LLM-compiled wiki to turn project documents into searchable, linked, and source-aware engineering knowledge.
Recoverable Agents in AgentScope Java 2.0: Event Streams, HITL, and Resume Semantics
A bilingual source code walkthrough of AgentScope Java 2.0 event streams, human approval, external tool execution, interruption, and session recovery.
AgentScope Java 2.0 Permissions: From ALLOW, ASK, and DENY to Sandbox Isolation
A bilingual engineering analysis of AgentScope Java 2.0 permission evaluation, HITL approval, tool validation, sandbox isolation, snapshots, and concurrency guards.
Recoverable agents, explicit permissions, long-term memory, and dependable product infrastructure.