Command Palette

Search for a command to run...

中文
“What looks like idleness may be the very moment when serendipity strikes.”
Fig. 1.
Yi Qiao's avatar

Yi Qiao

Building reliable AI agents.

Social links

Overview

Master of Computer Science student @University of Sydney

GitHub contributions

About

  • Hi! I'm Yi Qiao (Joey), a Master of Computer Science student at the University of Sydney focused on AI agent engineering.
  • I build multi-agent systems where control flow lives in code: LangGraph state graphs, typed tool gateways, Zod-validated boundaries, and deterministic fallbacks when a model fails.
  • I work agent-first every day with Claude Code, Codex and Cursor, and invest in the harness around them: Skills, AGENTS.md rules, CI checks and reproducible E2E verification.
  • I write about what I learn at qiao1.top (250+ posts). Available for an internship of six months or longer.

Experience

Shanghai Mingyao Information Technology

Operations and maintenance of legacy and new contract management systems (MySQL, Activiti workflow engine, Linux).

  • Traced contracts across legacy and new MySQL databases by joining contract, sign/payment form, business form and Activiti audit tables to locate the root cause of each ticket.
  • Repaired workflow data: reset stuck sign and payment forms, removed duplicated pending tasks, corrected approval timestamps, and fixed unpaid-amount and contract-term errors by recomputing payment aggregates.
  • Resolved login, approval and seal-authorisation failures in the RBAC model by auditing user–department–role mappings and adding missing organisation licences.
  • Watermarked replacement contract PDFs with a Java tool and deployed them over SSH/SCP in sync with database records; cleaned city data with Python and SQL; wrote a SQL runbook covering 10+ incident types.
  • MySQL
  • Activiti
  • RBAC
  • Java
  • Python
  • SQL
  • Linux

Beijing Jiuqi Software

Location
Ningbo, China
Location type
(On-site)
  • SQL
  • Data Reconciliation
  • Client Support

Projects(3)

  • Period
    09.2026—

    A multi-agent travel workspace: describe a trip in chat and get a validated plan with itinerary, transport, stays, dining, destination tips and budget, each item labelled with where its data came from. Plans are edited by chatting or directly on the day timeline and Google map. University team project; I built most of the system. Live demo.

    • Deterministic orchestration: a LangGraph state graph owns the flow (dispatch → conflict detection → targeted revision → assembly). A LangChain supervisor only picks which specialists each node calls, so control flow, budget limits and stop conditions stay in code.
    • Five specialist agents (itinerary, transport, accommodation, destination, dining) share one proposal board and run in stages: transport fixes routes and costs first, accommodation derives cities and nights from it, then itinerary and dining read what is left of the budget.
    • Bounded revision: budget and time conflicts re-run only the affected agents; revision stops when the plan score stops improving (max 3 rounds), and an infeasible budget reports the minimum needed instead of looping.
    • Reliability: Zod 4 validates brief, proposals and final plan at every graph boundary. If a model is unavailable or returns invalid output, rule-based parsing and deterministic fallbacks still complete the request.
    • Typed tool gateway: SerpApi and Google Places/Routes go through one gateway with quotas, a 15-minute cache and provenance labels (live, estimated or unpriced); a mock mode makes every flow reproducible offline.
    • Agent development harness: task-specific Skills (provider integration, UI verification, code review, pre-push checks), AGENTS.md rules shared across AI tools, decision records, and CI checks on protected files, shared contracts and docs, with E2E tests that keep reproducible evidence.
    • TypeScript
    • Next.js
    • LangGraph
    • LangChain
    • DeepSeek
    • Zod
    • Clerk
    • SerpApi
    • Google Maps
    • Turborepo
  • Period
    02.2025—
  • Period
    04.2026—

Side projects

Stack

Education

  • Relevant coursework: randomised and advanced algorithms, large-scale networks, advanced network technologies, information theory and self-organisation, model-based software engineering.
  • Advanced Algorithms
  • Large-scale Networks
  • Information Theory
  • Model-based Software Engineering
  • Machine Learning
  • Deep Learning
  • DSA
  • Computer Networks
  • Operating Systems

Recognition(3)

  • Type
    Award
    Prize
    2nd Prize
    Received in Grade
    University
  • National Encouragement Scholarship

    Type
    Award
    Prize
    Scholarship
    Received in Grade
    University
  • University Scholarship

    Type
    Award
    Prize
    Second-class
    Received in Grade
    University
Command Palette

Search for a command to run...