Tae-Eun Song
Managing director at Daejeon Jungang Cheonggua Co., Ltd. I work with wholesale-market and agricultural data through AI and code.
I spent about ten years in materials research at NNFC and KAIST. Around 2023 I started using AI tools seriously, and it shifted how I think about what's buildable. In 2024 I moved to Daejeon Jungang Cheonggua Co., Ltd.
Turning the problems I run into on the market floor into code, I ended up writing a few papers on AI verification and sending security patches to open-source libraries. Most days I'm somewhere between the field and the code — bumping into things, fixing them.
I also run a YouTube channel called "Sapjil Coding" (Korean for trial-and-error coding). The rule there is simple: only share what I've actually built and used, honestly. Everything I've made is gathered here in one place.
What I've made
Key results
- CCR · D-CCR — papers on cross-verifying code review with several AI models. arXiv: 2603.12123 · 2603.16244
- XMR — a unified writeup of the method, submitted to the TACL journal and currently under review (2026.05).
- HCCA — the same approach applied to medical coding, as a preprint: arXiv:2603.21454
- Union (medec) — combining several models' outputs caught medical-coding errors that a single model missed, in a medical-domain experiment.
- Open-source security — PRs to brotli / snappy / protobuf-c / jansson, plus 15 CVE reports (under review).
Areas
Wholesale & agricultural data
Working at Daejeon Jungang Cheonggua Co., Ltd., I build tools for handling the market's data.
Research & papers
Four papers so far, in areas I find interesting. The topics are covered in the papers themselves.
Open-source contributions
I contribute to open-source libraries. In 2026: 15 CVE reports and 4 PRs, under review.
Content · Sapjil Coding
I share what I've actually built and used, on video. No overstating it; if I don't know something, I say so.
Recent
- 2026.05 XMR — unified paper submitted to TACL (under review)
- 2026.05 GitHub profile README — results in one place
- 2026.05 brotli — size computation bounds PR
- 2026 HCCA preprint