Document Intelligence — Wealth Manager Guide

Extract structured data from PDFs and images in under 60 seconds. No manual data entry. No spreadsheet gymnastics.

What You Can Extract

Sentinel’s Nexus pipeline reads the following document types:

Document Type	Data Extracted	Typical Use
CAS Statements	Holdings, transactions, NAVs, XIRR	Portfolio reconciliation
AIF Statements	L1/L2/L3 holdings, capital calls, distributions	Alternative investment tracking
PMS Reports	Stock-level holdings, sector allocation	Discretionary mandate monitoring
MF Order Forms	Scheme, amount, mode, bank details	Order validation
KYC Forms	Client demographics, FATCA, nominee	Onboarding compliance

Step-by-Step: Upload → Review → Export

Step 1: Upload

Navigate to Documents in the Sentinel sidebar.
Drag and drop your PDF or click Upload.
Sentinel pre-classifies the document type automatically.
Click Process to start the 12-stage extraction pipeline.

Tip: For best results, ensure scans are 300 DPI or higher. Blurry images may trigger the review queue.

Step 2: Monitor Progress

The progress bar shows real-time stage updates:

Upload ✓ → Classify ✓ → Layout Detect ✓ → Extract ✓ → Validate → Store

Light documents (≤10 pages): ~30 seconds
Heavy documents (11–100 pages): ~2 minutes
Extra-large (>100 pages): ~5 minutes

Step 3: Review Extraction

Once complete, you’ll see:

Holdings table with ISIN, scheme name, units, NAV, market value
Confidence scores per field (green = high, amber = review, red = manual check)
Summary panel with total AUM, asset allocation breakdown, and top 5 holdings

Click any field to edit it. Your correction feeds back into the validation engine.

Step 4: Export

Format	Best For
Excel	Client presentations, CRM import, reconciliation
PDF Report	Formal client communication
JSON	System-to-system integration

Common Scenarios

Scenario: Multi-Account CAS

A single CAS contains 4 family accounts. Nexus detects and isolates each account automatically. You can:

View each account separately
Merge into a consolidated family view
Export per-account or consolidated Excel

Scenario: Handwritten Order Form

Upload a signed MF order form. The KimiOCR Agent (specialized handwriting model) reads the form and maps it to the order schema. Operations receives a notification for final approval.

Scenario: Batch Processing

Select up to 20 documents and click Batch Process. Documents queue automatically by size. You’ll receive a notification when the batch completes.

Troubleshooting

Issue	Cause	Resolution
“Low confidence on ISIN”	Poor scan quality or non-standard layout	Edit the field manually or re-upload at higher DPI
“Document classification failed”	Unsupported format or corrupted PDF	Convert to standard PDF and retry
“Validation error: sums don’t match”	Missing page in upload or table split across pages	Check page count; use “Stitch Table” option
“Extraction timeout”	Document >100 pages or complex tables	Document splits automatically; check Heavy queue

Best Practices

Upload original PDFs when possible — scanned images lose metadata
Verify page count after upload — missing pages are the #1 cause of validation errors
Review amber fields within 24 hours — they improve model accuracy for your tenant
Use consistent file naming — e.g., ClientName_CAS_YYYYMMDD.pdf
Export before end-of-day — extractions are retained for 7 years, but exports are client-deliverable

Document Intelligence for Wealth Managers