Update app.py

app.py

@@ -1,6 +1,6 @@
 from fastapi import FastAPI, File, UploadFile, HTTPException
 # Import Decimal and the custom JSONResponse
-from decimal import Decimal
+from decimal import Decimal, InvalidOperation
 from fastapi.encoders import jsonable_encoder
 from starlette.responses import JSONResponse
 import pytesseract
@@ -47,35 +47,71 @@ def convert_scientific_decimals(obj):
     elif isinstance(obj, list):
         return [convert_scientific_decimals(item) for item in obj]
     elif isinstance(obj, Decimal):
-        # Check if the decimal is in scientific notation by converting to string
-        decimal_str = str(obj)
-        if 'E' in decimal_str or 'e' in decimal_str:
-            # Convert to float first to get the actual numeric value, then back to Decimal
-            # This forces the decimal representation
+        # Always convert to proper decimal format, regardless of current format
+        try:
+            # Convert to float to get actual numeric value
             float_val = float(obj)
-            # Use high precision string formatting to avoid float precision issues
-            if abs(float_val) < 1e-10:  # Very small numbers
-                formatted = f"{float_val:.15f}"
-            elif abs(float_val) < 1e-6:  # Small numbers
-                formatted = f"{float_val:.12f}"
-            elif abs(float_val) < 1:  # Numbers less than 1
-                formatted = f"{float_val:.10f}"
-            else:  # Larger numbers
-                formatted = f"{float_val:.8f}"
             
-            # Remove trailing zeros and decimal point if not needed
-            formatted = formatted.rstrip('0').rstrip('.')
-            if '.' not in formatted and float_val != int(float_val):
-                # If we removed all decimal places but it's not actually an integer
-                # This handles edge cases
-                formatted = f"{float_val:.1f}".rstrip('0').rstrip('.')
+            # Handle special cases
+            if float_val == 0:
+                return Decimal('0')
+            
+            # For very small positive numbers, use high precision
+            if 0 < abs(float_val) < 1e-10:
+                formatted = f"{float_val:.20f}".rstrip('0').rstrip('.')
+            elif 0 < abs(float_val) < 1e-6:
+                formatted = f"{float_val:.15f}".rstrip('0').rstrip('.')
+            elif 0 < abs(float_val) < 1:
+                formatted = f"{float_val:.10f}".rstrip('0').rstrip('.')
+            elif abs(float_val) < 1000000:
+                formatted = f"{float_val:.8f}".rstrip('0').rstrip('.')
+            else:
+                # For large numbers, don't use decimal places if they're all zeros
+                if float_val == int(float_val):
+                    formatted = str(int(float_val))
+                else:
+                    formatted = f"{float_val:.2f}".rstrip('0').rstrip('.')
             
+            # Ensure we have at least some decimal representation for very small numbers
+            if formatted == '0' and float_val != 0:
+                formatted = f"{float_val:.20f}".rstrip('0').rstrip('.')
+                
             return Decimal(formatted)
-        else:
+        except (ValueError, OverflowError, InvalidOperation):
+            # If conversion fails, return the original
             return obj
     else:
         return obj
 
+def force_decimal_format(data):
+    """
+    Additional layer to ensure all numeric values are in proper decimal format.
+    This is applied right before JSON encoding.
+    """
+    if isinstance(data, dict):
+        result = {}
+        for key, value in data.items():
+            if key in ['unit_price', 'total_price', 'tax_amount', 'discount', 'net_amount', 
+                      'sub_total', 'tax_total', 'discount_total', 'total_amount', 'tax_rate'] and isinstance(value, dict) and 'value' in value:
+                # Special handling for numeric fields
+                if isinstance(value['value'], (Decimal, float, int)) and value['value'] != 0:
+                    try:
+                        float_val = float(value['value'])
+                        if 0 < abs(float_val) < 1e-6:
+                            # Very small numbers - use high precision
+                            decimal_str = f"{float_val:.15f}".rstrip('0').rstrip('.')
+                        else:
+                            decimal_str = f"{float_val:.10f}".rstrip('0').rstrip('.')
+                        value['value'] = Decimal(decimal_str)
+                    except:
+                        pass
+            result[key] = force_decimal_format(value)
+        return result
+    elif isinstance(data, list):
+        return [force_decimal_format(item) for item in data]
+    else:
+        return data
+
 # --- END OF MODIFICATIONS ---
 
 
@@ -176,12 +212,13 @@ async def process_with_gemini(filename: str, raw_text: str):
 - If a field is missing or not found, return an empty value (`""` or `0`) and set `accuracy` to `0.0`.
 - Convert any date found in format: YYYY-MM-DD
 
-CRITICAL: ALL numeric values must be in full decimal notation. NEVER use scientific notation:
+CRITICAL: ALL numeric values must be in full decimal notation. NEVER EVER use scientific notation or exponential format:
 - CORRECT: 0.0000009, 0.00000015, 0.0000002, 1500000, 0.00123
-- INCORRECT: 9e-7, 9E-7, 1.5e-7, 1.5E-7, 2e-7, 2E-7, 1.5e+6, 1.23e-3
-- For very small numbers like 0.0000009, write out all the zeros
-- For large numbers like 1500000, write out all the digits
-- This applies to unit_price, total_price, tax_amount, discount, net_amount, sub_total, tax_total, discount_total, total_amount
+- ABSOLUTELY FORBIDDEN: 9e-7, 9E-7, 1.5e-7, 1.5E-7, 2e-7, 2E-7, 1.5e+6, 1.23e-3, any number with 'e' or 'E'
+- For very small numbers like 0.0000009, you MUST write out all the zeros: 0.0000009
+- For large numbers like 1500000, you MUST write out all the digits: 1500000
+- This is MANDATORY for: unit_price, total_price, tax_amount, discount, net_amount, sub_total, tax_total, discount_total, total_amount
+- Example: if unit price is 9 * 10^-7, write it as 0.0000009, NOT 9e-7 or 9E-7
 
 Raw text:
 {raw_text}
@@ -261,8 +298,9 @@ Output JSON:
         # object inside Python, instead of a standard float.
         structured_data = json.loads(json_str, parse_float=Decimal)
         
-        # CRITICAL: Convert any scientific notation Decimals to decimal format
+        # CRITICAL: Multiple layers of conversion to ensure proper decimal format
         structured_data = convert_scientific_decimals(structured_data)
+        structured_data = force_decimal_format(structured_data)
         # --- END OF MODIFICATIONS ---
         
         structured_data_cache[text_hash] = structured_data
@@ -396,6 +434,9 @@ async def extract_and_structure(files: List[UploadFile] = File(...)):
     # 3. Use the custom encoder when returning the final JSON response.
     # This ensures the Decimal objects are converted to strings correctly.
     # The default JSONResponse will not handle Decimal types properly.
+    
+    # Apply final decimal format enforcement before encoding
+    output_data = force_decimal_format(output_data)
     encoded_data = custom_encoder(output_data)
     return JSONResponse(content=encoded_data)
     # --- END OF MODIFICATIONS ---