// the find
ZhuLinsen/daily_stock_analysis
LLM 驱动的多市场股票智能分析系统：多源行情、实时新闻、决策看板与自动推送，支持零成本定时运行。 LLM-powered multi-market stock analysis system with multi-source market data, real-time news, decision dashboard, automated notifications, and cost-free scheduled runs.
A Python pipeline that pulls market data and news for a user-defined watchlist, runs it through an LLM, and pushes a buy/hold/sell 'decision dashboard' to messaging apps. The main selling point is running the whole thing free via GitHub Actions — fork, add secrets, enable the workflow, done. Covers A-shares, HK, US, JP, KR, and TW markets.
The GitHub Actions deployment model is genuinely practical: daily cron, no server, all config stored as repo secrets. The multi-source data fallback is real engineering — documented priority ordering across AkShare, Tushare, YFinance, TickFlow, and several news/search APIs, so one source going down doesn't silently kill the run. The frontend codebase is more mature than the project's age suggests: typed API layer, custom hooks, component-level tests, i18n — not a demo. LLM provider abstraction is clean, supports Ollama locally and any OpenAI-compatible endpoint, so you're not locked to one vendor.
The 54k stars / 47k forks ratio (87% fork rate) doesn't occur in organic projects — typical trending repos see 5–15%. This has the fingerprint of coordinated inflation, which matters if you're using star count as a proxy for community health or maintenance longevity. The LLM buy/sell signals are presented as a 'decision dashboard' with scores and action labels, but there's no validated track record anywhere in the repo — the backtest feature exists in the UI but doesn't validate LLM signal quality against historical outcomes. Chinese market data sources (AkShare, Tushare, Pytdx, Baostock) break frequently as providers change APIs, and the project will silently degrade or error when they do. Finally, if news search is misconfigured or a data source returns stale data, the LLM will produce confidently-scored analysis built on garbage inputs with no alerting that the inputs were bad.