Agentic operations
Agents for constrained decisions
Exploring how agents can reason over operational constraints, coordinate tools, and keep people in control of consequential decisions.
Sydney, Australia · Open to collaboration
Engineering intelligence you can trust.
An independent agent practice by Yikun Yang—bringing together systems engineering, research, and university IT education. Every system is designed to keep reasoning explainable, actions traceable, and architecture ready to evolve.
Scroll to inspect the system
01 Selected work
Ideas made testable.
A selection of agent systems and research directions. Detailed case studies, screenshots, and public repositories will be added as project permissions allow.
PRIVACY /
Project descriptions are intentionally high-level for now. No private data, client details, or restricted implementation details are shown.
Multi-agent research
Coordination beyond a single model
Prototyping systems where specialised agents share state, negotiate responsibilities, and evaluate collective outcomes.
Learning technology
AI that supports learning, not shortcuts it
Designing tutor-style agents and feedback workflows that help learners reflect, practise, and progress while keeping educators in control.
02 Agent Lab
Watch an agent think in systems.
A small interactive trace showing one way an agent can move from intent to reviewed action. This is a front-end simulation today and a home for live experiments tomorrow.
Intent
Human input
Planner
Decompose task
Tools
Execute safely
Memory
Update context
Reviewer
Check outcome
Execution trace
Ready
01Capture intent and constraints
02Build a verifiable plan
03Call approved tools
04Store useful context
05Review before delivery
Interactive front-end simulation · no data leaves this page
03 About
Researcher. Engineer. Educator.
A teaching mindset, applied to agent systems: make the reasoning visible, the process traceable, and the architecture open to growth.
Yikun’s university IT teaching experience shapes the practice: complex ideas should remain understandable, feedback should be useful, and people should never lose sight of how an outcome was reached. That perspective carries through agentic workflows, multi-agent decision making, optimisation, and AI-assisted learning—connecting research depth with practical software.
01
Explainable by design
Reasoning people can inspect and understand
02
Traceable in action
Observable state, decisions, and tool activity
03
Built to extend
Modular agents, tools, memory, and workflows
04
University IT education
Teaching, feedback, and human-centred systems
04 Working notes
Questions worth exploring.
A future home for research notes, technical breakdowns, and lessons from building agents. These topics are currently marked as drafts rather than published articles.
What makes an agent workflow reliable?
Thinking beyond demos: observable state, bounded actions, evaluation, and graceful failure.
Memory is a product decision
How context, persistence, privacy, and forgetting shape an agent’s real behaviour.
Where should humans stay in the loop?
Matching oversight to risk, reversibility, uncertainty, and the cost of a wrong action.