Bhaskar2611's picture
Update app.py
bca7a79 verified
raw
history blame
14.6 kB
import os
import re
import json
import gradio as gr
import pandas as pd
import pdfplumber
import pytesseract
from pdf2image import convert_from_path
from huggingface_hub import InferenceClient
from fpdf import FPDF # Added for PDF generation
import tempfile # Added for temporary file handling
# Initialize with reliable free model
hf_token = os.getenv("HF_TOKEN")
client = InferenceClient(model="mistralai/Mistral-7B-Instruct-v0.2", token=hf_token)
def extract_excel_data(file_path):
"""Extract text from Excel file"""
df = pd.read_excel(file_path, engine='openpyxl')
return df.to_string(index=False)
def extract_text_from_pdf(pdf_path, is_scanned=False):
"""Extract text from PDF with fallback OCR"""
try:
# Try native PDF extraction first
with pdfplumber.open(pdf_path) as pdf:
text = ""
for page in pdf.pages:
# Extract tables first for structured data
tables = page.extract_tables()
for table in tables:
for row in table:
text += " | ".join(str(cell) for cell in row) + "\n"
text += "\n"
# Extract text for unstructured data
page_text = page.extract_text()
if page_text:
text += page_text + "\n\n"
return text
except Exception as e:
print(f"Native PDF extraction failed: {str(e)}")
# Fallback to OCR for scanned PDFs
images = convert_from_path(pdf_path, dpi=200)
text = ""
for image in images:
text += pytesseract.image_to_string(image) + "\n"
return text
def parse_bank_statement(text, file_type):
"""Parse bank statement using LLM with fallback to rule-based parser"""
# Clean text differently based on file type
cleaned_text = re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]', '', text)
if file_type == 'pdf':
# PDF-specific cleaning
cleaned_text = re.sub(r'Page \d+ of \d+', '', cleaned_text, flags=re.IGNORECASE)
cleaned_text = re.sub(r'CropBox.*?MediaBox', '', cleaned_text, flags=re.IGNORECASE)
# Keep only lines that look like transactions
transaction_lines = []
for line in cleaned_text.split('\n'):
if re.match(r'^\d{4}-\d{2}-\d{2}', line): # Date pattern
transaction_lines.append(line)
elif '|' in line and any(x in line for x in ['Date', 'Amount', 'Balance']):
transaction_lines.append(line)
cleaned_text = "\n".join(transaction_lines)
print(f"Cleaned text sample: {cleaned_text[:200]}...")
# Try rule-based parsing first for structured data
rule_based_data = rule_based_parser(cleaned_text)
if rule_based_data["transactions"]:
print("Using rule-based parser results")
return rule_based_data
# Fallback to LLM for unstructured data
print("Falling back to LLM parsing")
return llm_parser(cleaned_text)
def llm_parser(text):
"""LLM parser for unstructured text"""
# Craft precise prompt with strict JSON formatting instructions
prompt = f"""
<|system|>
You are a financial data parser. Extract transactions from bank statements and return ONLY valid JSON.
</s>
<|user|>
Extract all transactions from this bank statement with these exact fields:
- date (format: YYYY-MM-DD)
- description
- amount (format: 0.00)
- debit (format: 0.00)
- credit (format: 0.00)
- closing_balance (format: 0.00 or -0.00 for negative)
- category
Statement text:
{text[:3000]} [truncated if too long]
Return JSON with this exact structure:
{{
"transactions": [
{{
"date": "2025-05-08",
"description": "Company XYZ Payroll",
"amount": "8315.40",
"debit": "0.00",
"credit": "8315.40",
"closing_balance": "38315.40",
"category": "Salary"
}}
]
}}
RULES:
1. Output ONLY the JSON object with no additional text
2. Keep amounts as strings with 2 decimal places
3. For missing values, use empty strings
4. Convert negative amounts to format "-123.45"
5. Map categories to: Salary, Groceries, Medical, Utilities, Entertainment, Dining, Misc
</s>
<|assistant|>
"""
try:
# Call LLM via Hugging Face Inference API
response = client.text_generation(
prompt,
max_new_tokens=2000,
temperature=0.01,
stop=["</s>"] # Updated to 'stop' parameter
)
print(f"LLM Response: {response}")
# Validate and clean JSON response
response = response.strip()
if not response.startswith('{'):
# Find the first { and last } to extract JSON
start_idx = response.find('{')
end_idx = response.rfind('}')
if start_idx != -1 and end_idx != -1:
response = response[start_idx:end_idx+1]
# Parse JSON and validate structure
data = json.loads(response)
if "transactions" not in data:
raise ValueError("Missing 'transactions' key in JSON")
return data
except Exception as e:
print(f"LLM Error: {str(e)}")
return {"transactions": []}
def rule_based_parser(text):
"""Enhanced fallback parser for structured tables"""
lines = [line.strip() for line in text.split('\n') if line.strip()]
# Find header line - more flexible detection
header_index = None
header_patterns = [
r'Date\b', r'Description\b', r'Amount\b',
r'Debit\b', r'Credit\b', r'Closing\s*Balance\b', r'Category\b'
]
# First try: Look for a full header line
for i, line in enumerate(lines):
if all(re.search(pattern, line, re.IGNORECASE) for pattern in header_patterns[:3]):
header_index = i
break
# Second try: Look for any header indicators
if header_index is None:
for i, line in enumerate(lines):
if any(re.search(pattern, line, re.IGNORECASE) for pattern in header_patterns):
header_index = i
break
# Third try: Look for pipe-delimited headers
if header_index is None:
for i, line in enumerate(lines):
if '|' in line and any(p in line for p in ['Date', 'Amount', 'Balance']):
header_index = i
break
if header_index is None:
return {"transactions": []}
data_lines = lines[header_index + 1:]
transactions = []
for line in data_lines:
# Handle both pipe-delimited and space-delimited formats
if '|' in line:
parts = [p.strip() for p in line.split('|') if p.strip()]
else:
# Space-delimited format - split by 2+ spaces
parts = re.split(r'\s{2,}', line)
# Skip lines that don't have enough parts
if len(parts) < 7:
continue
try:
# Handle transaction date validation
if not re.match(r'\d{4}-\d{2}-\d{2}', parts[0]):
continue
transactions.append({
"date": parts[0],
"description": parts[1],
"amount": format_number(parts[2]),
"debit": format_number(parts[3]),
"credit": format_number(parts[4]),
"closing_balance": format_number(parts[5]),
"category": parts[6]
})
except Exception as e:
print(f"Error parsing line: {str(e)}")
return {"transactions": transactions}
def format_number(value):
"""Format numeric values consistently"""
if not value or str(value).lower() in ['nan', 'nat']:
return "0.00"
# If it's already a number, format directly
if isinstance(value, (int, float)):
return f"{value:.2f}"
# Clean string values
value = str(value).replace(',', '').replace('$', '').strip()
# Handle negative numbers in parentheses
if '(' in value and ')' in value:
value = '-' + value.replace('(', '').replace(')', '')
# Handle empty values
if not value:
return "0.00"
# Standardize decimal format
if '.' not in value:
value += '.00'
# Ensure two decimal places
try:
num_value = float(value)
return f"{num_value:.2f}"
except ValueError:
# If we can't convert to float, return original but clean it
return value.split('.')[0] + '.' + value.split('.')[1][:2].ljust(2, '0')
def process_file(file, is_scanned=False):
"""Main processing function"""
if not file:
return empty_df()
file_path = file.name
file_ext = os.path.splitext(file_path)[1].lower()
try:
if file_ext == '.xlsx':
# Directly process Excel files without text conversion
df = pd.read_excel(file_path, engine='openpyxl')
# Normalize column names
df.columns = df.columns.str.strip().str.lower()
# Create mapping to expected columns
col_mapping = {
'date': 'date',
'description': 'description',
'amount': 'amount',
'debit': 'debit',
'credit': 'credit',
'closing balance': 'closing_balance',
'closing': 'closing_balance',
'balance': 'closing_balance',
'category': 'category'
}
# Create output DataFrame with required columns
output_df = pd.DataFrame()
for col in ['date', 'description', 'amount', 'debit', 'credit', 'closing_balance', 'category']:
if col in df.columns:
output_df[col] = df[col]
elif any(alias in col_mapping and col_mapping[alias] == col for alias in df.columns):
# Find alias
for alias in df.columns:
if alias in col_mapping and col_mapping[alias] == col:
output_df[col] = df[alias]
break
else:
output_df[col] = ""
# Format numeric columns
for col in ['amount', 'debit', 'credit', 'closing_balance']:
output_df[col] = output_df[col].apply(format_number)
# Rename columns for display
output_df.columns = ["Date", "Description", "Amount", "Debit",
"Credit", "Closing Balance", "Category"]
return output_df
elif file_ext == '.pdf':
text = extract_text_from_pdf(file_path, is_scanned=is_scanned)
parsed_data = parse_bank_statement(text, 'pdf')
df = pd.DataFrame(parsed_data["transactions"])
# Ensure all required columns exist
required_cols = ["date", "description", "amount", "debit",
"credit", "closing_balance", "category"]
for col in required_cols:
if col not in df.columns:
df[col] = ""
# Format columns properly
df.columns = ["Date", "Description", "Amount", "Debit",
"Credit", "Closing Balance", "Category"]
return df
else:
return empty_df()
except Exception as e:
print(f"Processing error: {str(e)}")
return empty_df()
def empty_df():
"""Return empty DataFrame with correct columns"""
return pd.DataFrame(columns=["Date", "Description", "Amount", "Debit",
"Credit", "Closing Balance", "Category"])
# New function to generate PDF from DataFrame
def generate_pdf(df):
"""Generate PDF from DataFrame and return file path"""
if df.empty:
return None
# Create a PDF
pdf = FPDF()
pdf.add_page()
pdf.set_font("Arial", size=8) # Smaller font to fit more data
# Set column widths
col_widths = [22, 65, 20, 15, 15, 25, 20] # Adjusted to fit all columns
# Headers
headers = df.columns.tolist()
for i, header in enumerate(headers):
pdf.cell(col_widths[i], 10, header, border=1)
pdf.ln()
# Data
for _, row in df.iterrows():
for i, col in enumerate(headers):
# Truncate long descriptions
value = str(row[col])
if headers[i] == "Description" and len(value) > 30:
value = value[:27] + "..."
pdf.cell(col_widths[i], 10, value, border=1)
pdf.ln()
# Save to temporary file
temp_file = tempfile.NamedTemporaryFile(suffix=".pdf", delete=False)
temp_file.close()
pdf.output(temp_file.name)
return temp_file.name
# Modified Gradio Interface
with gr.Blocks() as interface: # Changed to Blocks for more control
gr.Markdown("## AI Bank Statement Parser")
gr.Markdown("Extract structured transaction data from PDF/Excel bank statements")
# File input
file_input = gr.File(label="Upload Bank Statement (PDF/Excel)")
# Output dataframe
output_df = gr.Dataframe(
label="Parsed Transactions",
headers=["Date", "Description", "Amount", "Debit", "Credit", "Closing Balance", "Category"],
datatype=["date", "str", "number", "number", "number", "number", "str"]
)
# State to store the processed DataFrame
state_df = gr.State(value=pd.DataFrame())
# Download button (initially hidden)
download_btn = gr.DownloadButton(
"Download as PDF",
visible=False,
elem_classes="download-btn"
)
# Process file and update state
def process_and_store(file):
df = process_file(file)
return df, df, gr.DownloadButton(visible=not df.empty)
# Connect components
file_input.change(
process_and_store,
inputs=[file_input],
outputs=[output_df, state_df, download_btn]
)
# Generate PDF when download button is clicked
def on_download_click(df):
return generate_pdf(df)
download_btn.click(
on_download_click,
inputs=[state_df],
outputs=[download_btn]
)
# Add custom CSS for the download button position
interface.css = """
.download-btn {
margin-top: 20px !important;
margin-bottom: 30px !important;
}
"""
if __name__ == "__main__":
interface.launch()