I am a student, and I currently make tools for myself that can process bulk personal text and recordings.
Long-term AI chats · Audio / video · Coding agents · Documents · Fully local
|
It can organize a person's long-term chat history across multiple AI platforms by day, and each conclusion can reference the original sentence and its timestamp. Fully local. From my own data, I got 128,000+ messages and 133,000+ evidence cards. It also has a fork (my classmate made it work on Windows and DeepSeek).
|
An open-source audio/video notebook that lets you ask across hundreds of hours of audio, and it will give you an answer similar to Google Gemini Notebook: when you click on a source, it will jump to the original second of the source (if it is a video, it will open at that second). It runs fully locally.
|
|
A self-written coding agent. It started from the mini-swe-agent tutorial as a learning reference; the routing, planning, context compression, file tools, sub-agents, and evaluation setup were written by me. It can read GitHub issues, modify the code, run the tests, and verify the fix. About 400 lines.
|
A project that can convert PDF, PPT, and Word into modifiable Markdown, and allows reverse conversion. If there is a text layer, it reads it directly; if not, it will be based on OCR. (It was driven by hard coursework during IB, and by current converters that were inconvenient for me to use.)
|
-2E5E4E){kind=icon}
I am recently contributing to the Open Notebook community (two open pull requests: a source-upload fix and zh-CN translation fixes), and the project has some parts (like its architectural rigor) for me to learn from.