Financial Document Intelligence

Document classification and extraction for financial documents with interpretability outputs.

Supported Document Types

Type	Examples
earnings_call	Quarterly earnings transcripts, investor day presentations
risk_filing	SEC 10-K/10-Q risk factors, annual report disclosures
trade_confirmation	OTC/exchange trade confirmations, ISDA confirmations
compliance_report	FINRA/SEC audit reports, remediation summaries

Quick Start

pip install -r requirements.txt
python examples/run_pipeline.py   # processes 3 sample documents
pytest tests/

Pipeline

Document Text
      │
      ▼
  Classifier ──→ doc_type + confidence + per-type scores
      │
      ▼
  Extractor ──→ dates, amounts, companies, CUSIP/ISIN, risk statements
      │
      ▼
  Interpreter ──→ plain-English explanation + key findings + risk signals
      │
      ▼
  PipelineOutput (all above combined)

Interpretability

Every classification decision produces a plain-English explanation:

• Why it was classified as this document type
• Which keywords drove the decision
• Confidence score with explanation
• Risk signals (breach/default/violation keywords)

No black box — every output is auditable.

Sample Output

Classified as: earnings_call (confidence: 85%)
Explanation: This document appears to be an earnings call transcript...
             Classification confidence: 85%. The earnings_call pattern
             matched 40% of its indicator keywords.
Key findings:
  - Financial amounts found: $4.2 billion, $1.87, $16.5 billion
  - Dates referenced: Q3 2024, October 15, 2024