Encouraging Kids to Read: LLMs for Education & Fun
summary
SoftwareOne builds the Book Aligned Activities AI Engine for Worldreader using AWS (S3, Lambda, Step Functions, Bedrock) and a Claude-style model to extract ePub text, run prompt-driven few-shot generation, postprocess and store JSON activities. It generates 10 activities per book across 16 categories (initially five: self‑awareness, social, physical, vocabulary, reading comprehension), the content reviewer picks the best three, and development time falls from weeks to minutes; delivered within a month and supports Kiswahili and English.
AI-written summary · about
chapters (16)
- 0:00Reading encouragement project intro
- 1:09World Reader’s literacy challenge
- 4:24BookSmart activity generation
- 6:15Solution design with LLMs & AWS
- 9:04Model selection & LM Arena eval
- 11:58System message for activities
- 13:23Few-shot & chain-of-thought
- 15:15Language consistency & compliance
- 16:39Generated activities examples
- 18:05Project outcomes & impact
- 19:34Key takeaways & zero scaling
- 21:22Q&A: Authors & engagement
- 23:12Q&A: Safety & human eval
- 25:41Q&A: Scaling beyond children’s books
- 27:26Q&A: Translating activities
- 29:33Closing remarks
related talks
24mBuilding Safe, Creative AI Experiences for Kids
34mBuilding Owly: An AI Comic Video Generator for My Son
39mBuilding and running an AI powered product
07mReading is the New Writing in the AI Era
05mKeynote: Teaching LLMs New Knowledge
49mNext-Gen Documentation with AI
28mA community based open source LLM and what it can do for your business
25mHow Mixbook leveraged genAI to offer personalized experiences