Production-grade Financial RAG for SEC 10-K filings with structure-aware parsing, semantic chunking, multi-stage retrieval, Jina reranking, and source-grounded answers

Natural language search for SEC filings. Fetch 10-K/10-Q/8-K from EDGAR, parse into segments, chunk text, generate GPU-accelerated embeddings (google/embeddinggemma-300m, 768-dim), store in ChromaDB+SQLite dual-store. Rich CLI with progress bars, color-coded output, flexible filtering. Returns relevant passages not generated answers.