Sangwon Ji
I'm an AI engineer who builds multi-agent systems that check their own work
Orchestration, retrieval over knowledge graphs and evals, shipped as services with tests, traces and one-command deploys.
- Based in
- Waterloo, ON
- Studying at
- University of Waterloo
- Working on
- Growing The Sovereign's learning pipeline and its review queueHardening Catchbunny's MCP endpoint for Claude Code and Codex
Experience
Full resumeQorsa
Machine Learning Engineer (AI Engineering)
Made report-writing agents 6× faster on 20× fewer tokens
- 83k → 4ktokens per report section
- 77s → 12sto write each section
- ~25ktokens saved per section, with tools run in code
Python, FastAPI, Pydantic, Dgraph, Langfuse
SK Networks & ENCO
AI Full-Stack Apprentice
Shipped real-time AI apps to AWS
- <1slive data over WebSocket and SSE
- 0manual deploy steps, with CI/CD on AWS
FastAPI, Redis, Next.js, OpenSearch, AWS
Selected work
Browser testers that prove a bug before they report it
- 9/9planted bugs found
- 78 → 28steps to reproduce, after shortening
- 4.6×faster with 8 testers than 1
- step
- detector fired
- happened again
- did not repeat
A research lab that learns every night and proves it on an exam
- 41 → 55held-out exam score after its research
- 1,010sourced facts from 47 pages in one night
- 9×faster fact extraction on a local model, same quality
- task
- report
- tool call
- facts stored
- dropped
Ask Claude about what you're reading, right from the MacBook notch
- ~4.4klines of Swift
- 60+built-in self-test checks
- 0servers, it runs fully on your Mac
Hiring an AI engineer? Let's talk
Open to AI engineering internships and roles. Email is the fastest way to reach me.