Refine AI Alignment in Northquill Platform
Employer not named by the sourceRemote
Frontier is not the employer and does not collect applications.
About this role
Animation, Voice Talent, Machine Learning (ML), Commercials, Video Editing, Natural Language Processing, AI Model Development, AI Integration · Hi — I have an existing SaaS product called Northquill.ai.
Northquill takes an existing PowerPoint presentation plus narration and creates a finished narrated MP4.
The particularly important use case is NOT AI-generated narration. It is taking one existing continuous long-form recording and automatically determining when the presentation should advance through the original slides.
My real-world stress test is approximately:
- 4 hours of continuous existing narration - roughly 150 PowerPoint slides - no existing timestamp map
Northquill already has an AI alignment/review workflow, but it consistently fails at this scale.
The biggest problems are:
- suggested timings lean heavily toward approximately equal splitting of the total audio across the slides; - the AI does not understand the semantic relationship between narration and individual slides well enough; - some slides may legitimately remain on screen for seconds while others may remain for several minutes; - the slide thumbnails in the existing review interface are currently failing to load; - the review step exists, but too many of the AI recommendations need correction.
I do NOT want a simple improvement to the equal-