InfinityUniverse0/ArXivToday-Lark

ArXivToday-Lark feels like a small project built to solve a real daily annoyance: keeping up with arXiv without manually checking the same categories every morning. The value is clear right away. You configure the arXiv categories and keywords you care about, optionally let an LLM narrow the results and translate abstracts, and then get a nicely formatted Lark/Feishu card in the team chat. For research groups or people tracking fast-moving AI/security topics, that is genuinely useful.

The code is also easy to understand. main.py has a straightforward pipeline: fetch papers, remove duplicates, filter them, translate when enabled, save history to papers.json, and post to Lark. Splitting the arXiv logic, LLM calls, webhook formatting, and config loading into separate files keeps the project approachable without overengineering it. The README is one of the stronger parts of the repo: screenshots, setup steps, bilingual docs, and an example config all make it much easier for someone to try.

The biggest thing I would improve is reliability. Right now there do not seem to be tests or CI, even though a few parts are easy to accidentally break: deduplication, keyword filtering, JSON history, and the Lark card payload. A small test suite would go a long way. I would also revisit keyword matching, because splitting abstracts by spaces can miss phrases like “intellectual property” or words next to punctuation. The webhook call should probably use a timeout and clearer retry/failure behavior too.

Overall, this is a practical, nicely scoped automation tool. It is not trying to be a big platform, and that works in its favor. With a bit more testing and operational hardening, it could become a dependable daily research assistant for Lark-based teams.