All Articles

How I Cut This Site's CLAUDE.md by More Than Half

Trimmed a 273-line CLAUDE.md down to 121 lines by separating templates into their own files and prioritizing guardrails.

Antigravity Caps Image Resolution at 1024×1024 Even with Gemini Pro

Even with an active Gemini Pro subscription, Nano Banana Pro isn't available in Antigravity yet — here's why image editing resolution is stuck at 1024×1024.

SeaArt LoRA Training — Practical Notes: Getting It to Generate as Intended

A hands-on log of three iterations of LoRA training on SeaArt. Summarizes model selection, parameter tuning, and lessons learned from failures.

Token Management Guide for Those Struggling with Bloated CLAUDE.md

A practical guide to getting Claude Code to build a service to a runnable state. Centered on token management, it organizes CLAUDE.md, skills, MCP, sub-agents, and verification strategies.

SeaArt LoRA Training Guide

How to train LoRA in the cloud on SeaArt and download a safetensors file for local generation.

LoRA: Setting Up a Training Environment on an RTX 3060 Laptop (6 GB VRAM)

Step-by-step guide to building a LoRA training environment on Windows 11 with an RTX 3060 Laptop (6 GB VRAM) using kohya_ss — from caption writing to VRAM-saving settings.

npm Was Broken So I Switched to pnpm [React2Shell Vulnerability]

npm install kept failing with a mysterious error while patching Next.js/React for a security vulnerability. Switching to pnpm fixed it immediately.

Notes on addressing the Next.js/React 'React2Shell' vulnerability

A summary of how to verify impact and the mitigation steps for the CVSS 10.0 React2Shell vulnerability (CVE-2025-55182 / CVE-2025-66478), plus additional DoS and source code exposure issues.

RAG: Building an Internal Help Desk with Mac mini M4 Pro + Dify (2025, Part 1)

A plan to build an internal help desk RAG system using a Mac mini M4 Pro and Dify. Highlights what's new in Dify circa 2025 and tips for running local LLMs.

Setting Up LLM/RAG for Work, So I Tried LoRA Training on the Side (Part 1)

Introducing a Mac mini M4 Pro to build an in-house RAG system. A plan for setting up a LoRA training environment during downtime while waiting for specs to be finalized.

Japanese OCR on the Web in 2025: Limits and Lessons

From browser OCR and server-side OCR to cloud APIs and AI — a roundup of what I learned trying to implement Japanese OCR on the web, including the limits of each approach.

@paddlejs-models/ocr Doesn't Work in Browsers (as of 2025)

A record of my failed attempt to use PaddleOCR's JavaScript implementation in the browser.