Slicing a 30,000px Webtoon Without Cutting Speech Bubbles in Half
A practical approach to splitting very tall images for OCR and translation, with code.
Oct 7, 20266 min read
Command Palette
Search for a command to run...
Latest articles
The Indie Tech Stack 2026: How I Ship High-Traffic AI and Privacy SaaS Solo with Next.js 14 and Edge Caching
From handling heavy AI computer vision pipelines to scaling global edge requests without a massive backend team: Here is the exact architecture I use to ship production-ready micro-SaaS apps solo. Whe
Sep 16, 20263 min read
Engineering a Seamless 2D-to-3D Mesh Pipeline: Overcoming the Geometry Bottleneck
Generative AI has evolved rapidly beyond text and 2D images, moving straight into spatial geometry. However, engineering a production-ready pipeline that transforms a flat 2D image into a structured,
Jul 9, 20263 min read
Under the Hood: Engineering a High-Throughput AI Video Translation & Voice Cloning Pipeline
Moving beyond basic API wrappers to solve context-aware video localization and native vocal fingerprint replication at scale. Hey developers, Video localization in 2026 is still fundamentally broken.
Jun 22, 20263 min read
Beyond Text: The Engineering Behind Seamless AI Manga Typesetting
Why translating comics is an image-processing challenge, not just a language one.
Apr 21, 20262 min read
Building a Manga Translator Taught Me Something Unexpected About Reading
When I first started building a manga translator, the goal seemed straightforward: Take Japanese text → translate it → show the result Like most developers, I approached it as a technical problem. Im
Apr 16, 20263 min read
Why I Stopped Relying on Generic LLMs for Manga Translation (And What I Built Instead)
The "Wrapper Trap" in AI software and the technical debt of ignoring spatial context in Japanese Manga.
Mar 31, 20263 min read
Low-Latency Image Processing in Next.js 14: Optimizing for Vertical Text Recognition
The Engineering Challenge Processing high-resolution images for optical character recognition (OCR) typically involves heavy server-side lifting, leading to significant latency. When dealing with spec
Mar 30, 20262 min readHow AI Manga Translators Work (OCR + Computer Vision + LLMs)
AI is getting really good at translating text. But translating manga is a completely different problem. Unlike normal text translation, manga has: text embedded inside images speech bubbles with irr
Mar 17, 20264 min read