Independent Agentic AI Developer
AI Evaluation · Technical AI Governance · Agent Reliability · Evidence-Grounded Systems
Based in Malaysia · Open to worldwide remote and international contract opportunities.
I build and evaluate AI-assisted systems with an emphasis on testable behaviour, evidence quality, traceability, failure handling, and controlled release decisions.
My current focus is UniAgent at CIKM 2026 — Task 2: Solving, where I am developing an independent system for document-based administrative decisions.
Current public-development result: 18/21 on the official Task 2 spot-check set.
Final hosted-model and hidden-test results are still pending, so I do not claim hidden-test generalization or unverified accuracy improvements.
- AI agent evaluation and regression testing
- Technical AI governance and assurance controls
- Evidence-grounded decision support
- Deterministic fallbacks and bounded model verification
- Auditability, traceability, and evaluation integrity
- AI-assisted software development with explicit test and release criteria
A multilingual publishing platform integrating Hugo, Notion-backed content, Cloudflare Workers, localization, accessibility checks, media-rights controls, and production-safety validation.
The repository separates content synchronization, candidate validation, deployment, and externally visible operations behind explicit validation gates.
An open working framework for making consequential or reusable AI workflows easier to inspect, challenge, validate, revise, and re-test.
Core workflow:
Assumptions → Evidence → Counter-evidence → Decision → Validation → Revision
The project explicitly separates framework design from empirical performance claims; benchmark superiority is not claimed without reproducible evidence.
An evidence-aware interactive Go learning prototype.
The project separates practice data, public test material, delayed evaluation, external review, and formal learning-effect claims. Engineering completion is deliberately not treated as proof of educational effectiveness.
A reproducible competition snapshot demonstrating how the same user request and approval can produce different outcomes when the governing authority state changes.
The project uses synthetic data, explicit provenance boundaries, controlled A/B comparison, and reproducible local tests. No real payment or production authorization is claimed.
For my ongoing UniAgent work:
- 18/21 official public development cases
- 62/62 full regression tests passed
- 11/11 deterministic robustness tests passed
- 18/18 bounded-verification mock tests passed
- Docker build and artifact validation passed
- Structured prediction/decision trace consistency tested
- Experimental changes isolated from the frozen candidate
- Hosted-model experiment currently blocked pending official endpoint access
Mock success is treated as architecture validation, not as evidence of accuracy improvement.
Before focusing on AI-assisted development, I worked in:
- secondary-school computer and visual arts education
- photography and video production
- institutional media and documentation
- visual communication design
I studied Visual Communication Design at National Yunlin University of Science and Technology, Taiwan.
I'm currently interested in worldwide remote roles involving:
AI Evaluation · AI Assurance · Technical AI Governance · Agent Evaluation · Responsible AI Controls · Agentic AI