The Quantum Ledger is a high-fidelity Retrieval-Augmented Generation (RAG) platform designed for real-time fiscal analysis. It transforms static financial documents—10-Ks, earnings transcripts, and management reports—into a dynamic, searchable intelligence graph.
By leveraging IBM Docling for structural parsing and Qdrant for vector orchestration, the Ledger provides grounded, source-cited insights into the performance of major tech entities like NVIDIA, Meta, Alphabet, Broadcom (AVGO), and TSMC.
A professional-grade dashboard for interactive querying and audit.
- Quantum Hits Sidebar: Real-time visibility into similarity scores and raw document chunks used for every answer.
- Entity Filtering: Toggle focus between specific companies or perform cross-sector "All" analysis.
- Session Persistence: Chat history and retrieval metadata are maintained throughout your research session.
The Ledger features a universal discovery layer for US-listed equities.
- Auto-CIK Discovery: Enter any ticker (e.g.,
AVGO,ARM,SMCI). The system automatically queries the SEC master directory to resolve Central Index Keys (CIKs) in real-time. - The Golden Trio Triage: Automatically identifies and retrieves the three most critical artifacts for fiscal grounding: 10-K (Annual), 10-Q (Quarterly), and 8-K (Current/Earnings).
- Universal Support: Native support for both Domestic (10-K/8-K) and Foreign Private Issuers (20-F/6-K).
A sophisticated ingestion pipeline designed to bypass legal "noise" and target high-value financial data.
- Exhibit Prioritization: The engine automatically hunts for Exhibits 99.1 and 99.2 (CFO Commentary and Press Releases) before the main filing body, ensuring the AI is grounded in executive strategy.
- Ghost Link Immunity: Intelligent whitelisting ignores SEC navigation "ghost links" (404s), locking onto actual
.htmand.pdfartifacts with 100% reliability. - Safety-Capped Chunking: Optimized
RecursiveCharacterTextSplitterlogic ensures tables are preserved while staying strictly within the 512-token limits of high-performance embedding models.
Most LLMs treat financial tables as a "word soup." By leveraging IBM Docling, the Ledger preserves the structural hierarchy of financial matrices, ensuring that complex data is interpreted with relational accuracy.
| Component | Technology |
|---|---|
| Interface | Streamlit |
| Search Service | SEC EDGAR API w/ BeautifulSoup4 (XML) |
| LLM | Claude 3.5 Sonnet (Anthropic) |
| Vector DB | Qdrant (Dockerized) |
| Parser | IBM Docling |
| Embeddings | BAAI/bge-small-en-v1.5 |
| Orchestration | Python 3.12+ |
Ensure you have Docker installed for the vector database and an Anthropic API key in your .env.
# Clone the repository
git clone [https://github.com/vasyloki/quantum-ledger.git](https://github.com/vasyloki/quantum-ledger.git)
cd quantum-ledger
# Set up environment
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
# Spin up the Qdrant engine
docker run -p 6333:6333 qdrant/qdrant
# Launch the Platform
streamlit run app.py- Enter a ticker in the sidebar (e.g.,
NVDAorAVGO). - Click 🔍 Scout Trio to see the latest 2026 filings.
- Click 📥 Ingest to Ledger on a specific filing. The system will automatically target the CFO Commentary (Ex 99.1) first.
- Grounded! Once you see the notification, your Ledger is updated with that company's specific fiscal context.
General LLMs often "hallucinate" by blending pre-training data with your documents. The Ledger utilizes Hard Grounding, physically restricting the LLM's context to only the specific data points retrieved from your private Qdrant vault.
- Audit-Ready Verifiability: Every response is tethered to a "Quantum Hit."
- Deterministic Retrieval: If the fact isn't in your Ledger, the system is instructed to state it doesn't know rather than "guessing."
- Relational Integrity: By preserving table structures, we prevent the "word soup" effect that leads to incorrect fiscal constant extraction.