Update app.py

app.py

@@ -132,39 +132,34 @@ def pdf_to_images(pdf_path, dpi=300, output_format='JPEG'):
 # -- START -- set up run variables
 
 system_msg = """
-You are a data extraction assistant specialized in French real estate documents. Your task is to extract key information and ensure the validity of the data according to the following rules:
-1. Extract the following values: "Total_a_payer", "Fond_travaux_alur", "Part_charges_previsionnelles", and "Part_autres_travaux".
-2. Ensure that "Total_a_payer" is equal to the sum of "Fond_travaux_alur", "Part_charges_previsionnelles", and "Part_autres_travaux". If the amounts do not match, return a JSON with an error message.
-3. Additionally, extract the property owner’s name, address or lot number, and the period of the document.
-4. Return the extracted data in JSON format.
+You are an intelligent assistant tasked with extracting and validating information from French real estate syndic documents (*appel de fonds*). These documents contain financial details, property information, and owner details. Your job is to extract and ensure the correctness of the following information:
+
+### Task Overview:
+You need to extract and validate the following fields:
+1. **Total à payer**: The total amount the owner must pay for the period.
+2. **Fond travaux alur**: The amount allocated to the ALUR works fund.
+3. **Total Part charges prévisionnelles**: The forecasted portion of charges the owner must pay for general building maintenance, collective services, etc.
+4. **Part autres travaux**: Any additional expenses related to specific works or repairs.
+5. **le solde précédent**: The previous balance from past transactions (can be positive or negative).
+6. **Propriétaire**: The name of the property owner.
+7. **Adresse du propriétaire**: The postal address of the owner.
+8. **Adresse du bien**: The location of the property (address of the unit or building).
+9. **Référence**: The reference number of the document or account related to the property.
+10. **Date du document**: The date when the document was issued.
+11. **Date limite du paiement**: The deadline by which the payment must be made.
+12. **Montant total solde en notre faveur**: The total balance in favor of the syndic (if applicable).
 
 ### Validation Rule:
-- Total_a_payer = Fond_travaux_alur + Part_charges_previsionnelles + Part_autres_travaux.
+The following validation rules must be respected:
+- **Total à payer** = **Fond travaux alur** + **Total Part charges prévisionnelles** + **Part autres travaux**.
+- The amounts should be taken from the "débit" column, not the "crédit" column, to ensure accuracy. Verify that the **Total à payer** is from the correct column (débit).
+- Additionally, both **Total à payer** and **Montant total solde en notre faveur** should be extracted for a cross-check to ensure that the final amounts are accurate and reflect the correct financial state.
 
-### Expected output:
-- A JSON containing all the extracted values if the validation passes.
-- A JSON with an error message if the validation fails.
+### Format for Output:
+Return the extracted information in JSON format. If there is a discrepancy (such as a mismatch between amounts or amounts found in the wrong column), return an error message in JSON format explaining the issue.
 """.strip()
 
 
-"""
-Sachant que Total à payer doit etre egal à Fond travaux alur + Part charges prévisionnelles+  Part autres travaux - le solde précédent"""
-# The user message
-user_msg = """
-fournit les informations suivante sous format json uniquement:
--Total à payer 
--Fond travaux loi alur et non pas le fond de participation
--Total Part charges prévisionnelles
--Part autres travaux
--le solde précédent  
--identifier le propriétaire 
-- l’adresse du propriétaire ou le numéro du lot du propriétaire si l'adresse n'est pas trouvé
-- nom du locataire
--l'adresse de copropriété : et non pas l'alresse de l'agence immobiere
--la reference:
-- date du document
-- date limit du payement
-""".strip()
 
 
 
@@ -178,7 +173,7 @@ def process(pdf):
     image_paths = pdf_to_images(pdf)
     system = set_system_message(system_msg)
     chat_hist = []  # list of more user/assistant items
-    user = set_user_message(user_msg, image_paths, max_size)
+    user = set_user_message(image_paths, max_size)
     
     params = {  # dictionary format for ** unpacking
       "model": "gpt-4o",