NTU Master of Computing (Applied Artificial Intelligence)

I build auditable AI systems for financial and research workflows, with a focus on financial agents, evidence-grounded evaluation, model validation, and reproducible ML.

• Accepted ACM HCOMP 2026 full paper, sole author: From Answers to Audit Opinions: Cost-Aware Expert Verification of Financial Artifacts from Tool-Using LLM Agents. The paper studies evidence-grounded verification of financial artifacts produced by tool-using LLM agents.
• spy-public-market-model-validation: public v0.2.1 SPY adjusted-close model-validation release using purged walk-forward folds, finance and ML baselines, QLIKE/RMSE, hash-verified artifacts, and a reproducible synthetic demo. It makes no trading or alpha claim.
• auditable-financial-agents: public v0.2.0 research-code companion implementing the paper's core research-label and evidence/materiality logic on deterministic synthetic examples; it is not the official HCOMP implementation or a full empirical reproduction.
• l40s-llm-bench: public v0.1.6 reproducible LLM inference harness with configuration capture, raw-result schemas, CI, manifests, and a truthful no-real-GPU-results boundary.

• Financial AI and professional agents
• Evidence grounding, provenance, and auditability
• Model validation, robustness, and failure analysis
• Reproducible ML research engineering

My earlier finance, accounting, audit-oriented data review, and risk-advisory work shaped an interest in materiality, traceability, controls, and reliable decision support.

• LinkedIn

Public repositories contain only claims and artifacts that pass a disclosure and reproducibility check.