google/langextract

no linked package yet

* Reddit bar shows a per-repo velocity proxy (raw score / 100); the score formula uses the corpus-normalized version so a single repo's bar may not match its contribution to the corpus-wide ranking.

LLM驱动的 A/H/美股智能分析器：多数据源行情 + 实时新闻 + LLM决策仪表盘 + 多渠道推送，零成本定时运行，纯白嫖. LLM-powered stock analysis system for A/H/US markets.

Once you have Claude Code wired up with the right connections — MCPs, APIs, local tools — the friction to try something new basically disappears. I saw Google’s open source LangExtract repo (LLM-based text extraction with source grounding) and said “let’s see if we can use it.” Ten minutes later, I was running it on SEC filings. The setup: I already had an MCP and API pipeline that pulls filing text from EDGAR. LangExtract sits on top of that and extracts structured information — business drivers, risk factors, guidance language — with the actual source text attached. That source grounding is the key. You’re not getting a summary the model made up. You’re getting extracted passages you can trace back to the filing. What makes this powerful is what comes next. I’m building skills — structured prompts for what to look for in a filing: business model, capital structure, management guidance, competitive positioning. These skills come from my own equity research process, which I’m documenting as a wiki. The end state is an agent that reads an SEC filing and extracts the information I actually care about, the way I would do it myself. The whole thing — from discovering a repo to having a working extraction pipeline integrated with my own research process — took an afternoon. That’s what happens when your tools are already connected.

Matched through a repo-specific project phrase query.