Mac text recognition read the Korean in my scanned commentary almost perfectly. The Greek, English names and footnotes were another story. Here is how free published texts and an AI check against the page photos fixed them, and what every method cost.
The vendor’s software for my book scanner was hundreds of megabytes and slow. My Mac already saw the scanner as a camera, so I built a small capture app with AI that morning. By evening it also flattened pages, read the text and made a searchable PDF: 279 pages.
My Book Scanner’s Software Was Too Heavy, So I Built a Scanning App With AI in One Day Read More »
My Logos screen translator took 18 seconds per commentary column. The translation itself was not the slow part. Measuring each step found four places where the app was simply waiting, and removing them brought it to 4.9 seconds.
From 18 Seconds to 4.9: Where My Screen Translator Was Really Waiting Read More »
With Claude, I built a small Mac app that translates the English commentary under my mouse in Logos into Korean. It went through four versions before lunch, and the change that mattered most had nothing to do with speed.
I Built a Screen Translator for My Logos Commentaries. The Real Problem Wasn’t Speed. Read More »
A Saturday-night check found a missing livestream, a stale lower third and a camera bridge that had been dead for a day. Fixing them ended with cameras that return on their own during the sermon and sermon slides built by script.
Cameras and Sermon Slides on Autopilot: One Saturday Night in a Small Church Booth Read More »
I replaced one do-everything AI chat with departments, each with a one-page charter, a secretary team wired to iPhone Reminders and Google Calendar, and a list of what the AI must never do.
I Stopped Asking AI to “Do Everything.” I Gave It Departments Instead. Read More »
A solo pastor moved the weekly church bulletin into one text file that also feeds Instagram, the church website and the livestream. What worked, what broke, and one silent failure.
I Replaced Our Church Bulletin Template with a Text File (and Everything That Broke) Read More »
My voice-driven ProPresenter playlist tool lost a song because the API returned 204 success while doing nothing. That failure, and how I ended up building song files by hand.
I was sure my wireless camera ran 4 frames late. A clap test showed the audio was late, from a 200ms offset I had eyeballed months ago. The measurement and two wrong fixes.
One command from my office wakes the booth PC; logon, ProPresenter, OBS and the camera bridge then start in order. Measured times, three traps, and the auto-login trade.
