Shane Tan.

AI EngineerSingaporeSMU Information Systems · 2027

AI systems that follow rules — and know when to bring people in.

I design multi-agent workflows that follow rules, pause for a person when they should, and keep their evidence beside every claim.

See selected workLinkedInGitHub
Artwork
Artwork Abstract, not a system diagram. The gate stands for the policy checks that run before specialist agents are called.
Internship
AI Solutions Intern · 2026
Studying
BSc Information Systems, SMU · exp. May 2027
Based in
Singapore
Focus
Multi-agent reliability, human review, evaluation

Selected work, in detail.

Each project separates what I built from what the team built.

No. 01 — Multi-agent AI · 2026

finAI — the coordination layer

How a rule-first Clarifier and a deterministic Orchestrator keep a team's specialist agents inside policy.

Team prototype · SMU IS4000 / SMU-X · wealth-advisory use case

Read the case study
Artwork
Artwork Abstract: one request, a gate, the agents beyond it. Not the system architecture.

Built by me

  • Rule-first Clarifier that turns requests into typed intents
  • Deterministic Orchestrator and plan admission
  • Single-use approval cards

Team system

  • Specialist agents, built by teammates
  • Test-mode specialist providers

Engineering decisionsMade in my coordination layer

  • Structured intent before planning
  • Policy-fixed plans, admitted deterministically
  • Withhold, don't substitute, when client context changes

No. 02 — Corporate onboarding · 2026

Corporate Onboarding

A recoverable case workflow: when evidence conflicts, the case pauses, is reviewed, and resumes from where it stopped.

Team proof of concept · industry-sponsored final year project · not production

Read the summary
Artwork
Artwork Abstract: sources converging at a checkpoint that can pause a case. Not the system architecture.

My contribution

  • Co-developed the team's proof of concept
  • Contributed to the review-and-recovery workflows

Team systemSix-person team

  • Registry lookup (mock/sandbox)
  • OCR-assisted extraction
  • Reconciliation and ownership analysis

Engineering decisionsTeam-level architecture

  • Typed blockers halt a named step when evidence conflicts
  • Reviewed cases resume from saved checkpoints
  • Mock integrations and mock submission
No. 03
SafeTrace

Local-first computer-vision evidence review that links findings to their source frames · internship prototype in a team repository

My contribution: React/FastAPI integration, local workflow hardening, the evidence-review experience and the hosted-frontend-to-local-backend connector.

Read the case study

Engineering approach

  1. i.

    Deterministic safeguards

    Rules stay the authority; models assist inside boundaries.

    Seen in finAI →
  2. ii.

    Human oversight by design

    Exceptions pause for a person instead of producing a substitute answer.

    Seen in Corporate Onboarding →
  3. iii.

    Evidence-driven development

    Findings link back to their source, and limits sit beside results.

    Seen in SafeTrace →

Additional work

Experience & education

Experience

  1. AI Solutions Intern

Education

  1. BSc Information Systems, Singapore Management University (expected May 2027)
  2. Diploma in Aerospace Electronics, Singapore Polytechnic
Skills and credentials

Let's talk about reliable AI.

The best way to reach me is LinkedIn.

Connect on LinkedInGitHub