Scolastica — AI Educational Content Generator
Turning a source PDF into a polished, template-matching PowerPoint deck used to be entirely manual: reading every page and rebuilding slides by hand could take most of a working day. Scolastica turns that into an automated pipeline that still keeps a human in control of the final choice for every section.
An operator uploads a source PDF plus a PowerPoint master template. Claude proposes several layout variants per section, but nothing ships on trust: every plan is validated against the master's real placeholders — an invalid layout, a placeholder that doesn't exist, or an empty text fill triggers one automatic repair round-trip back to Claude instead of a silent fallback — and every text fill is scored against the source PDF for how much it actually overlaps with it, flagging anything that looks invented instead of shipping it quietly. The operator picks the best variant per section plus an image, and python-pptx assembles a final deck that matches the template exactly. What used to take most of a working day by hand comes down to about 30 minutes end to end.
The whole generation runs as a background job with an explicit, persisted state machine (queued → planning → grounding → rendering → building → completed), so a server restart mid-job doesn't silently lose it — a real problem the first request/response version had. The same pipeline extends to other content built from the same source material: auto-generated subtitles for audio/video, quizzes, and interactive maps.
Built entirely solo — design, backend, frontend, and the whole content-generation pipeline.