Data Flow Diagram

This document describes the end-to-end data flow through Smartflow, from document upload to LoanIQ export.

---

High-Level Flow

flowchart LR
    A["📄 Document Upload"] --> B["🔄 Processing Engine"]
    B --> C["🤖 AI Extraction"]
    C --> D["✅ Human Review"]
    D --> E["📤 Export to LoanIQ"]

    subgraph "Smartflow Platform"
        B
        C
        D
    end

    style A fill:#eef2f8,stroke:#2f588b
    style B fill:#ecf8fb,stroke:#1d92a9
    style C fill:#ecf8fb,stroke:#1d92a9
    style D fill:#ecf8fb,stroke:#1d92a9
    style E fill:#dbf5e6,stroke:#2e9d58

---

Detailed Flow

flowchart TD
    Upload["1. Document Upload
    PDF/DOCX credit agreement
    uploaded via browser or API"]
    -->
    Ingest["2. Document Ingestion
    Format validation, OCR if scanned,
    page segmentation, structure detection"]
    -->
    Extract["3. AI Extraction
    200+ field extraction using
    loan-specific AI models"]
    -->
    Confidence["4. Confidence Scoring
    Each field scored 0-100%.
    Fields ≥95% auto-verified.
    Fields <95% flagged for review."]
    -->
    Review["5. Human Review (HITL)
    Operations officer reviews
    flagged fields, confirms or corrects.
    Evidence links shown for each field."]
    -->
    Verify["6. Verification Complete
    All required fields verified.
    Document marked ready for export."]
    -->
    Export["7. Export to LoanIQ
    JSON payload sent via integration API.
    Field-level mapping applied.
    Reconciliation report generated."]
    -->
    Confirm["8. Confirmation
    Export status confirmed.
    Audit trail recorded.
    Document archived."]

    style Upload fill:#eef2f8,stroke:#2f588b
    style Ingest fill:#eef2f8,stroke:#2f588b
    style Extract fill:#ecf8fb,stroke:#1d92a9
    style Confidence fill:#ecf8fb,stroke:#1d92a9
    style Review fill:#fff2d6,stroke:#d18a08
    style Verify fill:#dbf5e6,stroke:#2e9d58
    style Export fill:#dbf5e6,stroke:#2e9d58
    style Confirm fill:#dbf5e6,stroke:#2e9d58

---

Data at Each Stage

Stage	Data Created	Storage	Retention
Upload	Original document (PDF/DOCX)	Document store (encrypted at rest)	Per retention policy
Ingestion	OCR text, page structure	Processing cache	Cleared after extraction
Extraction	Structured field values + evidence links	Extraction database	Permanent
Review	Review decisions, corrections, timestamps	Audit log	Permanent
Export	Export payload, reconciliation report	Export history	Permanent

info

All data resides within the bank's infrastructure. No document content is transmitted externally at any stage.

---

Error Handling

flowchart TD
    Error["Error Detected"]
    Error --> UploadErr["Upload Error
    Invalid format, file too large"]
    Error --> ExtractErr["Extraction Error
    Model timeout, corrupt page"]
    Error --> ExportErr["Export Error
    LoanIQ unreachable, validation failure"]

    UploadErr --> RetryUpload["User re-uploads
    with corrected file"]
    ExtractErr --> RetryExtract["Auto-retry (3x).
    Fallback: manual extraction."]
    ExportErr --> RetryExport["Auto-retry (3x at 30s intervals).
    Fallback: manual export trigger."]