Skip to main content

Where I am, autumn 2026

· 4 min read
Kit Kong
Kit Kong
Engineer in Vancouver, AI systems and trading infrastructure

The last post here is from December 2024. This one fills the gap: a master's degree, a few systems I built and ran, and what I want to work on next.

The degree​

In January 2025 I started an MSc in Applied Computer Science at FDU Vancouver, and I finished it in August 2026 with a CGPA of 3.56.

My capstone was about LLM serving. Our team of four built a learning-to-rank request scheduler on top of vLLM, and I owned the KV-cache layer underneath it. I wrote up what we measured and what surprised us in the capstone post. A related side project was getting a 27B model to run usefully on a single RTX 3090, which is in its own post.

Solana MEV infrastructure, 2025​

Most of my 2025 engineering time went into Solana MEV infrastructure. The piece I am proudest of is a Jito bundle relay. I first wrote it in Go, then ported it to Rust. It fans bundles out to 8 block-engine regions over more than 4,000 egress IPs, with token-bucket rate limits and cooldowns when a region pushes back.

Most of the work was about throughput and staying inside rate limits, not about the trading itself. The details are in the relay post.

A Rust trading system on Polymarket​

From late 2025 to the summer of 2026 I worked on a Rust trading system for Polymarket's 15-minute BTC up/down markets. I wrote about 80% of it, 720 of the 894 commits. It covers the market-data feed, signals, execution, a risk engine, reconciliation and gasless claims. It traded live through the spring and summer; a four-week on-chain census covered 895 of its markets.

Building it taught me more about backtests than about markets. I wrote that down in what a live trading system taught me about backtests.

Robinhood Chain arbitrage​

I also built arbitrage tooling for Robinhood Chain, an Arbitrum L2. I built it detection-first: measure the opportunities, then build the execution path. I explain why in measure before you build.

LLM-Digest​

Since May 2026 I have been running LLM-Digest, a self-hosted market-intelligence pipeline. It reads 127 sources and has produced more than 4,100 reports so far. The main report is a nine-stage chain where each stage reads its own material and passes its conclusions to the next stage. I describe how that is built in nine stages, one conclusion.

RootData and CatcherVC​

From September 2022 to September 2026 I was a co-founder of RootData, a crypto data platform. I designed and collected its database from zero, then ran product. Over that time it grew to 428K users and more than 200 API partners. Near the end we open-sourced MCP servers so AI agents could query RootData directly.

I was also an investment partner at CatcherVC, where I reviewed more than 200 seed-stage deals a year. That work trained a habit I still use in engineering: read the evidence before the pitch.

How I work now​

Most of my 2026 work was built with coding agents in the loop. I have strong opinions about what makes that work on a large codebase and what does not. I collected them in working with coding agents.

What I am looking for​

I am looking for full-stack, data or AI-systems roles in Vancouver. I am also interested in low-latency work. I like jobs where the data has to be right and the system has to stay up, and where I can work across the stack from the feed to the UI.

The case studies are on /work and the full history is on /resume. If you are hiring, or know someone who is, email me at [email protected].