I built Zanki to help my son (4 YO) practice his letter sounds and learn to read starter words. Other apps felt overproduced and more like entertainment than education. I wanted something that struck the right balance between fun and learning and collectible stickers has been working well. There’s lots of directions this can go, I’ve experimented with other types of cards (colors, shapes, numbers, etc.) and want to go deeper on personalization and tailoring the content to child’s interests on any given day. This type of content generation would be impossible without LLMs!
Tech: Zanki is written in Elixir and deployed to fly.io. It’s a Phoenix LiveView app and works best on a mobile browser (technically it can be bookmarked as a PWA but the xp needs work!). Content: Replicate + Huggingface + OpenAI are doing a bunch of the content gen heavy lifting. I used multi-step AI pipelines to generate all of the text content, the image prompts, images, and text to speech. I’m working on fine-tuning a story generation model to get better results.
