Current milestone
MergeBench reproduction with LoRA specialists: plan and code written, first GPU run still pending.
Ray's Lab
A public build journal for AI-powered learning tools like AntiCopilot and a Japanese learning app. Build logs, experiments, and essays from ongoing work.
Now: part-time research assistant, after a summer 2026 internship at IIS, Academia Sinica · 4th-year CS, NCKU
Current work
Current milestone
MergeBench reproduction with LoRA specialists: plan and code written, first GPU run still pending.
Recent experiment
Origin × host 2×2: adapters trained on the base model applied to the instruct model and the reverse, three seeds each, plus one adapter trained on the mixed data as the control for merging.
Current question
When two merge operators score differently, how much of that is the operator and how much is the total norm of the merged update?
Active projects
Summer 2026 research internship at the Institute of Information Science, Academia Sinica, continued part-time: once a LoRA adapter is trained, how much of it can be reused, on another model or merged with other adapters?
A highly personalized IDE-integrated learning experience powered by LangGraph, designed to help developers learn by doing — not by copying.
Featured
I'm not a data viz expert. Here are the prompt additions I use to get chart output past vanilla matplotlib, plus a recent experiment aligning visuals to my blog's style.
Why this piece
A good entry point if you're new here - more honest than a landing page, and closer to the actual work than a portfolio write-up.
Recent writing
The site moved to blog.nhade.com in May but kept pointing at its old address. Fixed, and there is an RSS feed.
AWS EC2's free credits ran out. Here's the choice tree I walked, and the €4/mo stack I landed on.
Why retrieval over textbook grammar is mostly noise, and what works instead — embedding the learner's own history of wrong answers.
Rethinking AI assistants by engineering a VS Code extension that uses spaced repetition to teach concepts instead of just autocompleting code.
About
I'm Ray, a 4th-year CS student at NCKU building AI-powered learning tools. After a summer 2026 research internship at the Institute of Information Science, Academia Sinica, I'm continuing that work as a part-time research assistant. This site is where I document what I'm building, what I'm learning, and the design decisions along the way.