EarSphere

EarSphere is the companion system to a research paper by Zilin Li, Shaofei Shen and Zhilong Xie (GitHub ZilinLi-Daniel). Designed for visually impaired online learners, it recommends news based on your interests and reading history, then uses a large language model to write summaries that are easier to listen to. The README says it relies on GPT-series APIs; the actual summary code points to a third-party compatible endpoint with a Claude model name. It's a research prototype with no hosted version, so you deploy it yourself.

Who it's for:Researchers and developers working on accessible reading who want to study this design. Readers who don't write code won't get much from it.

Before you start: You deploy a Vue 3 frontend, a Flask backend and a MySQL database yourself, with Python 3.7 or later, Node 14 or later, npm, and a GPT-compatible API account, which may cost money. The README is in English, and the backend includes Chinese and English strings. The backend handles usernames, emails, passwords, language and interests, stores them in MySQL, and logs reading history. On sign-up and login the code prints the email and password to the program log, and no password hashing was found in the app code; a set of access credentials is also hard-coded in the source. When it writes summaries, the full news text and your reading-history titles go to the third-party endpoint set in the code. If you use these features, weigh the safety risks yourself. AI Inspiration World is not responsible for any consequences. The repo has no license file, so permission to copy or use it commercially isn't stated.

Based on the project's public documentation. Not tested by AI Inspiration World.