AI
The Text of My Videos Was Invisible to My AI
By Khai Tran · · 2 min read
Fifty-five of my recordings had already been turned into text. The memory search my AI relies on was built to skip all of it.
Week ending October 16.
On Wednesday I asked Red to save and log the written text of my processed videos. It turned out the text already existed, and Red's own memory search was built to skip it.
I had run my recordings through the reel pipeline: Coaching Corner sessions, trainings, talking videos. The pipeline cut clips and, along the way, turned the speech into text. That text was stored as raw material, which the memory search Red relies on leaves out by design. For the 14 Coaching Corner recordings Red checked, nothing tracked pointed to their transcripts.
I asked for the same treatment for every processed video, and every future one. The count came to 61 recordings across nine batches. Fifty-five had transcripts. Six never did, because someone else was presenting or the screen was private.
Red wrote a script to rebuild the index, with instructions to run it after each new batch, made a separate copy of the transcripts, and added a tracked map with outlines, clip hooks and file paths but no transcript text. The rule for the usage log is one row whenever a recording feeds a written piece, to help catch accidental reuse of the same passage.
Quotes need a context check, and doubtful words need checking against the recording. These are machine transcripts, not read line by line. Third-party course recordings are excluded from the library and from content sourcing.
As of Wednesday, the usage log has no rows. The change that points the blog and quote card tools at the library is open and not merged.