Lecture Video Generation Pipeline
A PDF of study material in, a narrated twenty-minute lecture video out.
Problem
Turning written study material into a narrated lecture is hours of manual work per topic: writing a script, recording narration, building slides, assembling video. The inputs and outputs are both structured enough to automate.
Approach
A staged media pipeline with no database, because the job has no state worth persisting between runs. Text extraction from the PDF, a language model writing a slide-by-slide teaching script, speech synthesis for narration, slide rendering, and assembly into a finished video. Each stage produces a checkable artifact, so a failure is localised rather than requiring the whole run again.
Outcome
A finished narrated lecture video of roughly twenty minutes generated from a source PDF, in Arabic or English.
How it's built
Staged pipeline with artifacts between stages and no persistent store. Stateless by design, since the only durable output is the video itself.
No database
A pipeline whose input is a file and whose output is a file has no state to keep. Adding a database would create an operational surface for no benefit.
No run history or resumability, the correct trade for a job measured in minutes.
Adjacent systems
School Timetabling System
Conflict-free timetable generation across grades, sections, teachers and subjects, using a language model constrained by a hard specification and checked by real code.
Offline-First Field Operations Platform
Delivery drivers work for hours with no connectivity, then sync. The hard problem is not the sync, it is what happens when the field and the office disagree.
Offline-First Pharmacy Platform
A pharmacy management system that keeps selling when the internet drops, then reconciles every till operation exactly once when it returns.