Skip to main content

About me

I'm Kit, an engineer in Burnaby, BC. I came to code from finance: a business degree, then years of reading crypto deals and building a data product, and the systems I build now still start from what the numbers are for.

The short version

I studied finance at Simon Fraser University and spent the years after it in crypto: researching deals as an investment partner at CatcherVC, advising a family office, running overseas business development for ChainCatcher, and building co-founding RootData, whose database I designed and collected from zero, into a platform with 428K users.

In 2025 I went back to school for an MSc in Applied Computer Science and spent the same year building Solana MEV infrastructure. In 2026 I built LLM-Digest, an AI analyst I run every day, and finished a capstone on KV-cache LLM serving. The long version is in Where I am, autumn 2026.

I'm looking for AI, full-stack and data engineering roles in Vancouver. I work in English, Mandarin and Cantonese.

What I'm good at

  • AI systems

    LLM pipelines with typed stages and fallback models, retrieval with citations, model layers that outlive any one provider, and the serving side underneath: KV cache, kernels, local inference.

    See the work

  • Low-latency and on-chain

    Rust, Go and Python trading infrastructure: a Jito bundle relay across eight regions, a live Polymarket system, arbitrage on an EVM layer-2, measured before it is built.

    See the work

  • Product and data

    Co-founder of a crypto data platform that reached 428K users, with a database I designed and collected from zero, and four years reading 200+ seed deals a year. I know what the data is for before I write the pipeline.

    See the work

  • Infrastructure

    A Proxmox homelab with a Synology NAS, Linux and Arch on bare metal and in VMs, Docker and Kubernetes, machines configured with Ansible, and servers and address ranges across regions for the Solana relay. Production apps run on Vercel and Cloudflare.

    See the work

  • Building with AI

    Claude Code, Codex, OpenCode and Perplexity every day, with notes and checks that keep agents honest on large codebases. LoRA training and ComfyUI pipelines on a local RTX 3090.

    See the work

How I work

Habits that show up across the projects, each with the place it came from.

  1. Every number traces back

    LLM-Digest hands typed conclusions from stage to stage and links each figure to the post it came from. The payment indexer keeps an evidence row behind every address it attributes.

  2. Say what was observed and what was guessed

    The Solana analyses tag each claim as observed, inferred or unknown, with the file, RPC result or transaction that backs it. One command recomputes every table offline.

  3. Measure the result you were hoping not to get

    A send path cut from 672 ms to about 12 ms still landed none of 22 attempts from a remote host. That isolated distance, not code, as the limit, before more time went into code.

  4. Fail closed

    When a venue's fee turned out to be dynamic and unbounded, the router stopped routing that pool rather than price it with a guessed constant.

Background

Finance first, then engineering. The full history is under Experience.

  1. 2025 – now

    Independent quant and MEV developer

    Rust, Go and Python trading systems

  2. 2025 – 2026

    MSc, Applied Computer Science

    Fairleigh Dickinson University, Vancouver

  3. 2024 – 2026

    Investment advisor

    J17 Crypto, family office

  4. 2022 – 2026

    Co-founder, product manager and advisor

    RootData, crypto data platform

  5. 2022 – 2026

    Investment partner

    CatcherVC

  6. 2022 – 2026

    Overseas operations and business development

    ChainCatcher

  7. 2021

    BBA, Finance

    Beedie School of Business, Simon Fraser University

Write to me

Email [email protected], or find me on GitHub and LinkedIn. I'm on Pacific time.