Solo founder, writing from Ouagadougou. Built Readium — an AI-native reading platform that lets users "talk to characters" from classic novels via Claude Sonnet 4.5.
Along the way I kept hitting walls no AI tutorial covers:
- Streaming SSE through your own server (not the SDK)
- Parsing structured LLM output when some comes back malformed
- Caching LLM extractions so the bill doesn't 100x on launch day
- Holding an AI character in role across 10+ messages
Pulled the patterns into an 83-page field manual based on Readium's production code — drafted with Claude, edited and verified by me.
File paths, real mistakes, real fixes.
→ $29 ebook
→ $59 ebook + companion repo (Readium's source, MIT)
→ Free chapter 1 (SSE streaming) on the product page https://limack.gumroad.com/l/ai-reading-apps
Honest disclosure: I'm a novice founder. I designed Readium and verified what ships, Claude did the engineering. If you want technical depth, the chapters have file paths and real code — that's where it lives, not in me.
My other product is MediLearn, an edtech for francophone medical
students with paying users in West Africa.
What patterns would you want covered in a v2?
https://limack.gumroad.com/affiliates