Structured Data Extraction Pipeline

Content

{
  "steps": [
    {
      "tool": "llm",
      "order": 1,
      "action": "Classify each document by type: invoice, contract, report, email, other",
      "output": "doc_type_map"
    },
    {
      "tool": "llm",
      "order": 2,
      "action": "Apply type-specific extraction schema: route invoice→invoice_extractor, contract→contract_extractor, etc.",
      "output": "extracted_fields"
    },
    {
      "tool": "python_pydantic",
      "order": 3,
      "action": "Validate extracted fields: type check, required field presence, cross-field consistency",
      "output": "validation_report"
    },
    {
      "tool": "python_script",
      "order": 4,
      "action": "Flag low-confidence extractions (LLM confidence < 0.85) for human review queue",
      "output": "review_queue"
    },
    {
      "tool": "db_client",
      "order": 5,
      "action": "Write validated records to destination schema (PostgreSQL/BigQuery/CSV)",
      "output": "loaded_records"
    },
    {
      "tool": "python_script",
      "order": 6,
      "action": "Generate extraction report: success rate, field-level accuracy, review queue size",
      "output": "extraction_report"
    }
  ],
  "tools_required": [
    "llm",
    "python3",
    "pydantic",
    "db_client"
  ],
  "expected_output": "Structured records in target schema + extraction report + human review queue for low-confidence items",
  "trigger_condition": "Batch of unstructured documents (PDFs, HTML, emails) requiring data extraction"
}