YLYana Liinyanalibuilds.hashnode.dev·4h ago · 11 min readThe Model Exposed Uncertainty. My Subtitle Editor Hid It.In one test clip, the speaker said HP, but the auto subtitles showed gp. The rest of the transcript looked clean, so that one wrong acronym was easy to miss. The speech-recognition result had already 00
YLYana Liinyanalibuilds.hashnode.dev·Jul 14 · 5 min readA Finished AI Video Can Still Be a Failed Product StateA generation job can complete successfully while the product outcome has already failed. I learned that while building a faceless Shorts workflow: the MP4 existed, but the video still was not somethin00
YLYana Liinyanalibuilds.hashnode.dev·Jul 3 · 4 min readScript-first captions fail lessA common subtitle architecture mistake is to treat speech recognition as the source of truth even when the user already has an approved script. I used AI assistance to draft and edit this post, then r00
YLYana Liinyanalibuilds.hashnode.dev·Jul 1 · 7 min readAI Summaries Need ReceiptsDisclosure: I used AI assistance for editing and clarity. The product decisions, examples, and implementation lessons come from my own work building AudienceCue. When I was running YouTube channels, s00
YLYana Liinyanalibuilds.hashnode.dev·Jun 30 · 6 min readAI Memory Is Not a Prompt TrickI used AI to help edit this cross-post adaptation, but the product decisions, product boundaries, and examples are mine. The memory problem became real for me when I stopped thinking about preferences00
