app.py

import torch
from transformers import DistilBertTokenizer, DistilBertModel
import spacy
from typing import Dict, List, Tuple
import sqlite3
from datetime import datetime
import uuid

class AccountingNLP:
    def __init__(self):
        # Use DistilBERT for lower memory usage
        self.tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')
        self.model = DistilBertModel.from_pretrained('distilbert-base-uncased')
        # Load spaCy model
        self.nlp = spacy.load("en_core_web_sm")
        self.accounting_rules = self.load_accounting_rules()
        # Use /app for SQLite to ensure write permissions
        self.connection = sqlite3.connect('/app/accounting_db.sqlite')
        self.create_tables()

    def load_accounting_rules(self) -> Dict:
        """Load GAAP/IFRS rules and chart of accounts"""
        return {
            'accounts': {
                'cash': {'type': 'Asset', 'normal_balance': 'Debit'},
                'office_supplies': {'type': 'Asset', 'normal_balance': 'Debit'},
                'accounts_payable': {'type': 'Liability', 'normal_balance': 'Credit'}
            },
            'rules': {
                'purchase': {
                    'debit': ['office_supplies'],
                    'credit': ['cash', 'accounts_payable']
                }
            }
        }

    def create_tables(self):
        """Create database tables for general ledger and journal entries"""
        cursor = self.connection.cursor()
        cursor.execute('''
            CREATE TABLE IF NOT EXISTS journal_entries (
                entry_id TEXT PRIMARY KEY,
                date TEXT,
                account TEXT,
                debit REAL,
                credit REAL,
                description TEXT
            )
        ''')
        cursor.execute('''
            CREATE TABLE IF NOT EXISTS general_ledger (
                account TEXT,
                balance REAL,
                last_updated TEXT
            )
        ''')
        self.connection.commit()

    def process_input(self, text: str) -> Dict:
        """Process natural language input and extract intent and entities"""
        doc = self.nlp(text)
        entities = {
            'amount': None,
            'account': None,
            'date': None,
            'payment_method': None
        }

        # Extract entities
        for ent in doc.ents:
            if ent.label_ == "MONEY":
                entities['amount'] = float(ent.text.replace('$', ''))
            elif ent.label_ == "DATE":
                entities['date'] = ent.text
            elif ent.text.lower() in self.accounting_rules['accounts']:
                entities['account'] = ent.text.lower()

        # Simple intent classification
        intent = 'record_transaction' if 'record' in text.lower() else 'query'

        return {'intent': intent, 'entities': entities, 'raw_text': text}

    def generate_journal_entry(self, processed_input: Dict) -> List[Dict]:
        """Generate double-entry journal entries"""
        if processed_input['intent'] != 'record_transaction':
            return []

        entities = processed_input['entities']
        entry_id = str(uuid.uuid4())
        date = entities['date'] or datetime.now().strftime('%Y-%m-%d')
        entries = []

        if entities['account'] in self.accounting_rules['rules']['purchase']['debit']:
            # Debit entry
            entries.append({
                'entry_id': entry_id,
                'date': date,
                'account': entities['account'],
                'debit': entities['amount'],
                'credit': 0.0,
                'description': processed_input['raw_text']
            })
            # Credit entry (assuming cash payment for simplicity)
            entries.append({
                'entry_id': entry_id,
                'date': date,
                'account': 'cash',
                'debit': 0.0,
                'credit': entities['amount'],
                'description': processed_input['raw_text']
            })

        return entries

    def validate_transaction(self, entries: List[Dict]) -> Tuple[bool, str]:
        """Validate journal entries for double-entry compliance"""
        total_debit = sum(entry['debit'] for entry in entries)
        total_credit = sum(entry['credit'] for entry in entries)

        if total_debit != total_credit:
            return False, "Debits and credits must balance"
        if not entries:
            return False, "No valid entries generated"
        return True, "Valid transaction"

    def update_ledger(self, entries: List[Dict]):
        """Update general ledger with validated entries"""
        cursor = self.connection.cursor()
        for entry in entries:
            cursor.execute('''
                INSERT INTO journal_entries (entry_id, date, account, debit, credit, description)
                VALUES (?, ?, ?, ?, ?, ?)
            ''', (
                entry['entry_id'],
                entry['date'],
                entry['account'],
                entry['debit'],
                entry['credit'],
                entry['description']
            ))

            # Update general ledger balance
            cursor.execute('''
                INSERT OR REPLACE INTO general_ledger (account, balance, last_updated)
                VALUES (?, 
                    (SELECT COALESCE((SELECT balance FROM general_ledger WHERE account = ?), 0) 
                     + ? - ?),
                    ?)
            ''', (
                entry['account'],
                entry['account'],
                entry['debit'],
                entry['credit'],
                entry['date']
            ))

        self.connection.commit()

    def generate_response(self, processed_input: Dict, entries: List[Dict]) -> str:
        """Generate natural language response"""
        if processed_input['intent'] == 'record_transaction':
            is_valid, message = self.validate_transaction(entries)
            if is_valid:
                self.update_ledger(entries)
                return f"Successfully recorded transaction: {processed_input['raw_text']}"
            return f"Error: {message}"
        return "Query processing not implemented"

    def process(self, text: str) -> str:
        """Main processing pipeline"""
        processed_input = self.process_input(text)
        entries = self.generate_journal_entry(processed_input)
        return self.generate_response(processed_input, entries)

# Example usage
if __name__ == "__main__":
    accounting_ai = AccountingNLP()
    result = accounting_ai.process("Record a $500 office supplies purchase paid by check")
    print(result)