smartdoc classifier

The SharePoint Document Classification System is an intelligent automation tool that:

Automatically classifies documents in SharePoint based on content and filename patterns Applies metadata tags to organize and categorize documents Uses AI to generate classification rules from natural language descriptions Extracts text from images and scanned PDFs using OCR technology Calculates confidence scores using weighted analysis of document sections

How to Process Documents Method 1: Process with a Single Rule When to use: When you want to apply one specific rule to your documents.

Steps: 1. Go to the Process page 2. Under "Single Rule Processing", select a rule from the dropdown 3. Click "Process with Single Rule" 4. Wait for processing to complete 5. Click "View Results" to see what happened

Example: Select "Invoice Detection" to find and tag all invoices.

Method 2: Process with All Rules When to use: When you want the system to evaluate all rules and apply the best matches.

Steps: 1. Go to the Process page 2. Under "All Rules Processing", click "Process with All Rules" 3. Wait for processing to complete (this may take longer) 4. Click "View Results" to see what happened

Example: Use this when uploading mixed documents and want automatic categorization.

Understanding Results Success Messages ✓ Processed: 15 files ✓ Tagged: 12 files ✗ Failed: 3 files Processed: Total files examined Tagged: Files that matched rules and were updated Failed: Files that couldn't be processed (check errors) File Results Table Each row shows: - Filename: The document that was processed - Status: Success or failure - Tags Applied: Metadata added to the file - Confidence: How sure the system is (for all rules processing)

Confidence Scores (All Rules Processing) 90-100%: Very confident match 70-89%: Good match 50-69%: Moderate match Below 50%: Weak match (may need review) Supported File Types Documents PDF: Regular PDFs and scanned PDFs (with OCR) Word: .docx, .doc files Excel: .xlsx, .xls files Text: .txt files Images (OCR will extract text) JPEG/JPG: Photos and scans PNG: Screenshots and graphics GIF: Animated and static images BMP: Bitmap images TIFF: High-quality scans WebP: Modern web images

Microsoft Azure is used here primarily through Azure OpenAI Service to power the natural-language-to-YAML rule generation part of the solution. The flow is basically:

The user writes a plain-English description of a classification rule, and the application sends that prompt to an Azure OpenAI model deployment. Azure OpenAI then converts that description into a properly formatted YAML rule file that can be reviewed, edited, and saved into the existing rules system.

More specifically, Azure is used for:

Hosting the AI model: The solution requires an Azure OpenAI resource in Azure, along with a deployed model such as GPT-4 or GPT-3.5-turbo. Secure configuration and access: The app connects to Azure OpenAI using Azure-specific settings in the .env file, including: AZURE_OPENAI_ENDPOINT AZURE_OPENAI_API_KEY AZURE_OPENAI_DEPLOYMENT_NAME AZURE_OPENAI_API_VERSION Rule generation API calls: When the user clicks Generate Rule, the application calls the backend endpoint that uses Azure OpenAI to analyze the description, extract keywords and logic, choose the rule type, and generate valid YAML. Costed AI inference: The document also notes that usage is billed through Azure OpenAI API calls, based on model type and tokens consumed.