Evidence

Capability Highlights

Much of my client work is confidential, so this page focuses on evidence you can verify or inspect: shipped systems at commercial scale, competitive ML results, public code and writing, and the technical foundations behind my judgment.

  • 01 Competitive ML

    Kaggle Peak Global Rank #63

    Competing against thousands of practitioners sharpened my instincts for validation design, error analysis, and data leakage detection. I focus on validation pipelines that mirror production conditions, so models optimize for real-world generalization rather than leaderboard or validation-set overfitting.

    • Validation Design
    • Error Analysis
    • Feature Engineering
  • 02 ML Engineering

    Core ML Infrastructure for a $400M+ Online Retailer

    Led a small engineering team designing and shipping core ML components for a category-leading online retailer with $400M+ in annual revenue. I operate where statistical modeling meets production constraints, keeping systems resilient, low-latency, and maintainable.

    • ML Infrastructure
    • Systems Design
    • Resilient Pipelines
  • 03 Open Source

    Public Code & Technical Writing

    I maintain public GitHub work across utility tools, model reproductions, and occasional contributions to upstream libraries such as PyTorch Lightning. It offers a direct view into my coding standards, documentation habits, and communication style. I also write technical posts that explain papers and core ML concepts for a broader peer community.

    • Code Craftsmanship
    • Technical Communication
    • Community Contributions
  • 04 Foundations

    Computer Science & Statistics

    Formal computer science and statistics training in Taiwan and Australia grounds the way I reason about models and systems. I design clean software abstractions around mathematically sound assumptions, avoiding both brittle data contracts and unnecessary model bloat.

    • Applied Statistics
    • Clean Abstractions
    • Data Contracts

The profiles below provide lightweight verification. They are not a substitute for a project-specific review, but they show public work samples that technical reviewers can inspect directly.

Kaggle profile GitHub profile Technical blog

Targeted proof

Evidence follows the problem.

A static public portfolio rarely answers the questions that matter for complex ML work. A data-quality audit, a production inference bottleneck, a custom recommender system, and an evaluation methodology review each require different proof of competence.

Once we define your technical situation, I can share context-specific materials: anonymized technical memos, past architecture sketches, relevant code patterns, or references where appropriate. For teams ready to evaluate fit, we can initiate a small Paid Test Run using a bounded slice of your actual data or systems.

Ready?

Let's see if the problem fits.

Start a conversation →