LectureTrace

Turn local lecture recordings into a workspace you can trace back to the source.

简体中文 · Setup · Contributing · Security

Founder and primary maintainer: Harry (@zhouziyueharry-droid).

LectureTrace is a local desktop-browser application for studying recorded lectures. Watch the video alongside subtitles, slides, notes and PDFs; follow timestamps and citations back to the material when checking an answer. The current interface is primarily Simplified Chinese.

What it does

• Organize local MP4 recordings into courses.
• Generate timestamped subtitles with optional local Whisper ASR; translate subtitles with DeepL, Google, an API model, or local Ollama.
• Extract slides from a user-confirmed region and generate a timestamped outline.
• Ask questions with source references through an API model or an optional Codex runtime.
• Edit Markdown notes, read and annotate PDFs, and work in a resizable reading workspace.
• Export and import processed lectures or complete courses as .ltcourse packages.

The backend uses Python, FastAPI and SQLite. The frontend uses React, TypeScript and Vite. Your recordings and database stay in a local data directory. Local storage does not mean every feature is offline: API translation, slide reading, outline generation and model Q&A send selected content to the configured provider. Check the send plan and estimated costs before confirming.

Quick start

Install uv and Node.js 22.18+ (Node 24 LTS recommended). Python 3.12 is pinned in backend/.python-version; uv can install it. Install FFmpeg and ffprobe for video/audio processing.

git clone https://github.com/zhouziyueharry-droid/LectureTrace-OSS.git
cd LectureTrace-OSS
uv sync --directory backend --locked
npm ci --prefix frontend
npm --prefix frontend run build
uv run --directory backend --no-sync python -m app.main

Open http://127.0.0.1:8000. No API key or .env file is needed to launch the UI.

The default installation is lightweight: it does not install Whisper, CUDA or OCR models. It can manage notes/PDFs and read imported processed course packages. Processing a new recording needs optional ASR plus your chosen services; unavailable capabilities produce an explicit error. See full setup for CPU/GPU ASR, local translation, PDF OCR and developer mode.

Contribute

Issues and pull requests are welcome in English or Chinese. You do not need collaborator access: fork this repository, create a branch, and open a PR. Start with a reproducible bug, a focused usability improvement, documentation, or a test.

uv run --directory backend --no-sync pytest -q
npm --prefix frontend test
npm --prefix frontend run build

Tests use temporary data and mocked model services. Never add keys, recordings, course slides or private datasets to a PR. See CONTRIBUTING.md for attribution, review and validation expectations.

NTU AI6130 group members (26S1, Large Language Models): please read the group collaboration notes for our assignment workflow, access after the course, continued participation and contributor credit.

Status and limitations

This is an early public source release of an existing personal project. The public history starts with a reviewed source snapshot; private lecture materials and historical experiment artifacts are not included. Future contributions will be recorded through public commits and PRs.

• Designed for one trusted local user; not a hosted multi-user service. Keep the server on loopback.
• Model answers and extracted measurements can be wrong. Verify against the source material.
• Provider model availability and prices vary. Built-in estimates are dated approximations, not billing guarantees.
• Optional GPU, OCR and real provider calls need your own setup. Automated tests do not establish their end-to-end quality.
• PDF rendering includes large assets; the production build may report bundle-size warnings.

Architecture · Release validation · Changelog

License

LectureTrace project code is available under the MIT License, including commercial use. Dependencies and optional model weights retain their own licenses; see third-party notices. The project license does not grant rights to recordings, slides, PDFs or other content you import.