US-stock skilled-LLM Agents
ScannerDashboard
GitHub ↗

Task 2 — SEC 10-K Item Extractor

Paste any EDGAR 10-K URL. Layered pipeline: L1 anchor → L2 structural → L3 LLM self-consistency → quarantine if confidence stays low. Each layer only runs when cheaper layers fall short — most filings finish at L1 with $0 LLM cost.

What works / what doesn't ↗

Type anything — the system parses with a cheap LLM call (~$0.0001). Examples: Apple 2024JPM 2023 10-K微軟 年報Tesla, or a full EDGAR URL. Foreign filers (20-F) and non-10-K forms (10-Q etc.) get a typed refusal rather than wrong data.