Two voice memos (12:19 and 12:58) — Dan is sketching a business model for selling AI-generated graded reader EPUBs.
The idea
A website that:
- Sells short graded reader stories for ~€1 each
- Covers all languages, all CEFR levels, all topics
- If a language/level combo doesn’t exist yet, the user clicks a button → it generates → they get the EPUB
- Open source the prompt if people want to DIY
- Stories accumulate in a catalog so they don’t regenerate
The anti-slop mechanism
Dan is worried about quality repetition. His solution: a trope/premise database (like the Lute vocab database tracks word frequency) that tracks which narrative tropes, genres, and character archetypes have been used. This prevents the AI from recycling “evil genius has a redemption arc” endlessly.
Source premises from real writers and Reddit (per Dan’s existing anti-slop feedback) rather than AI-generic pitches.
Analysis
The graded reader skill already exists and produces high-quality output with Olly Richards benchmarks. The business extension needs:
- A web frontend (simple: catalog + buy/generate)
- A trope database (new)
- Payment processing (Stripe, simple)
- Automated generation pipeline (the skill already does this)
The market is real: language learning apps are a $101B market in 2026, and digital graded readers are outselling print. Apps like Readle and Eppika charge subscription fees — Dan’s per-story model is simpler and more accessible.
Key risk: “AI slop” perception. Dan’s anti-slop database idea is the differentiator.
Connected to: Dan’s graded reader skill, [[energy-and-software-economics]]
You must log in or # to comment.