ai-mistake-notebook: Android mistake notebook with analysis via DeepSeek Vision

⚡ The gist in 5 seconds - A project called XXYxiexieyu/ai-mistake-notebook has been published on GitHub — an Android app for working through mistakes, with image analysis via DeepSeek Vision. - The claimed features include mapping tasks to textbook knowledge, detecting corrections across the entire page, and AI-powered breakdown of each individual question; it was tested on a Xiaomi 15. - It must be installed manually from the repository — no Google Play builds are mentioned in the post. ### 🔍 What was found On GitHub, the XXYxiexieyu/ai-mistake-notebook repository has been published — an Android app of the wrong-question notebook class, i.e., a digital notebook for reviewing mistakes. The description lists four features: local-first data storage, vision-based image analysis via DeepSeek, textbook knowledge mapping (matching tasks to textbook knowledge), full-page correction detection (finding corrections across the whole page), and per-question AI Q&A — a separate dialogue with the model for each question. The author notes that the project has been tested on a Xiaomi 15. Here's how it works: the user takes a photo of a completed assignment page, and the app analyzes the image not fragment by fragment but across the entire spread — this way it catches edits and marks anywhere on the sheet — then breaks the material down into individual tasks and links them to the relevant textbook sections. After that, you can ask the model a question about each item and get an explanation. The local-first prefix in the description means the mistake notebook itself is stored on the device, not in a cloud account. ### 💡 Why it matters For students and tutors, the bottleneck isn't recording mistakes — it's reviewing them afterward: photographing a page isn't enough; you need to understand exactly what was corrected and which topic it relates to. The combination of "full-page correction detection + textbook mapping + Q&A on a specific question" addresses precisely this step, while local-first removes the main fear of parents and schools — sending children's work to third-party servers. The limitation is obvious too: the project is distributed as a repository, so installation is manual, and the vision part relies on DeepSeek, i.e., an external model. The source gives no timeline for app store availability or a stable build.