% grep -r "tech" /posts/
Tech
Auto-Push URLs to Bing Search Engines with IndexNow
Adding IndexNow to push new and updated articles to Bing instantly — no waiting for sitemap discovery. GitHub Actions handles the ping on every push.
Seeing "Tokuryū" Through a Systems Engineer's Eyes
A look at Tokuryū (Japan's anonymous, fluid crime groups) through an engineer's lens — not a pyramid, but a CaaS platform hacking gig-work to externalize risk.
Why I Skipped rehype-slug for Heading Anchors
Why this blog uses manual remark-heading-id over the standard rehype-slug for heading anchors — multilingual deep links, URL aesthetics, info design.
Front Matter CMS and Keystatic side-by-side: where their design philosophies clash
How combining Front Matter CMS and Keystatic exposed deep design philosophy differences — Markdoc round-tripping, IndexedDB drafts, and SPA cache pitfalls.
Deploying Keystatic to Vercel Preview
Deploying Keystatic to Vercel Preview — switching storage modes by NODE_ENV, blocking /keystatic in production, GitHub App setup, and the default-branch trap.
Trying Keystatic as a CMS for Local Content Editing
Trying Keystatic on a Next.js blog — local setup, isolating the admin UI with Route Groups, and the Markdoc configuration gotchas I ran into.
Setting Up a Markdown Blog Writing Environment with Front Matter CMS
How I set up Front Matter CMS in VS Code to manage my Markdown blog -- frontmatter field design, automation pipeline coexistence, and the operational rules I landed on.
Pagefind Japanese Search Deep Dive: Tokenizers and IME Fixes
Why searching "サイバー" (cyber) in Pagefind returns "サイドバー" (sidebar) -- digging into the tokenizer mismatch between the index side (lindera/IPAdic) and query side (Intl.Segmenter), plus fixing IME input issues.
Automating Related Posts with Embeddings
How I built an automated related-posts feature using Supabase's pgvector and semantic similarity instead of traditional keyword matching, with thoughts on hybridizing with other recommendation techniques.
My Translation Scored 0.96, But the English Version Fell Apart
A high back-translation score doesn't guarantee native-level English. I ran into a case where a 0.96-scoring translation was completely broken for native readers — and learned that translation quality has two orthogonal axes.
Automating Blog Tagging with AI
To cut through the hassle of tag granularity and reconciliation, I fully automated tag management using the Gemini API and GitHub Actions. Notes on how to embed rules in prompts and get stable automated classification working in the pipeline.
Testing 3 Different Translation Prompt Patterns
I tweaked the translation prompts to get more natural output — and accidentally polluted my own evaluation mechanism. Here's what I found.
AI Translation Model Comparison: Back-Translation Scores Across the Gemini Family
Comparing back-translation scores across multiple AI models including Gemini and Claude. Looking at how translation accuracy varies by article content, embedding model characteristics, and the difference between translate and evaluate modes.
Automated Translation Quality Verification with AI
I built a system that automatically verifies translation quality using back-translation — no reference translations needed. Scores are calculated with Gemini Embedding + pgvector cosine similarity, comparing results across multiple models.
Hello World
First post on a blog for movie and anime thoughts, tech notes, and whatever else comes up. Also covering the technical side — AI-powered multilingual support with automated translation quality checks across multiple models.