added SOP

prompts.py

@@ -1,46 +1,46 @@
-
 def get_ocr_extraction_prompt(raw_text: str) -> str:
     """
-    Returns a strict prompt with SOP and One-Shot example (City of Auburn invoice).
+    Constructs a prompt that forces the LLM to output JSON strictly adhering 
+    to the Zoho Invoice API Schema.
     """
+    
     return f"""<|im_start|>system
-You are a specialized Invoice Extraction Agent. 
-Your task is to extract structured data from OCR text into a CANONICAL JSON format.
-
-### STANDARD OPERATING PROCEDURE (SOP):
-1. **Sanitization**: Ignore page headers, footers, and marketing messages.
-2. **Dates**: Convert all dates to YYYY-MM-DD format.
-3. **Null Handling**: If a field is not found, set it to null. DO NOT invent data.
-4. **Line Items**: Extract the table of goods/services accurately.
-5. **Output**: Return ONLY valid JSON. No Markdown. No commentary.
-
-### ONE-SHOT EXAMPLE:
-**Input OCR**:
-"CITY OF AUBURN... Account Number: 076248-000... Due Date: 01/07/25...
-Water Total $649.69... Sewer Total $1,333.45... Total New Charges $2,363.39"
+You are an intelligent Invoice Processing Agent responsible for Data Entry.
+Your task is to extract data from the provided OCR text and format it strictly according to the **Target API Schema** below.
 
-**Correct JSON Output**:
+### TARGET API SCHEMA (Zoho Books):
+Output a single JSON object with these exact keys:
 {{
-    "invoice_number": "076248-000",
-    "vendor_name": "City of Auburn",
-    "invoice_date": "2024-12-18",
-    "due_date": "2025-01-07",
-    "total_amount": 2363.39,
-    "currency": "USD",
+    "customer_name": "string (The Vendor or Bill To entity)",
+    "invoice_number": "string",
+    "date": "string (YYYY-MM-DD)",
+    "due_date": "string (YYYY-MM-DD)",
+    "reference_number": "string (PO Number if available, else null)",
+    "total": float (The grand total amount),
     "line_items": [
-        {{"description": "Water Total", "amount": 649.69}},
-        {{"description": "Sewer Total", "amount": 1333.45}},
-        {{"description": "Storm Water Total", "amount": 380.25}}
+        {{
+            "name": "string (Description of goods/service)",
+            "quantity": float (Default to 1.0 if missing),
+            "rate": float (Unit price),
+            "item_total": float (Line amount)
+        }}
     ]
 }}
-<|im_end|>
 
+### STANDARD OPERATING PROCEDURE (SOP):
+1. **Analyze**: Read the OCR text below, accounting for potential noise or scanning errors.
+2. **Normalize Dates**: Convert all dates (e.g., 'Jan 15, 2024') to ISO format (2024-01-15).
+3. **Table Extraction**: Identify the line items table. If the OCR layout is messy, use the price and quantity columns to align rows.
+4. **Validation**: Ensure 'item_total' equals 'quantity' * 'rate'.
+5. **Output**: Return ONLY the JSON object. Do not include markdown formatting (```json) or conversational text.
+
+<|im_end|>
 <|im_start|>user
-### CURRENT TASK:
-Extract JSON from the following OCR text:
+### INPUT OCR TEXT:
+{raw_text[:4000]}
 
-{raw_text[:4000]} 
+### RESPONSE (Valid JSON):
 <|im_end|>
 <|im_start|>assistant
 {{
-""" # Pre-fill brace to force JSON mode
+""" # Pre-filling the brace forces the model into JSON generation immediately.