Spaces:
Runtime error
Runtime error
Add verification functionality and improve UI
Browse files
app.py
CHANGED
|
@@ -25,6 +25,12 @@ import pdfplumber
|
|
| 25 |
|
| 26 |
pdf_examples_dir = './pdfexamples/'
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
model = dspy.OpenAI(
|
| 29 |
model='gpt-3.5-turbo-0125',
|
| 30 |
api_key=os.getenv('OPENAI_PROJECT_KEY'),
|
|
@@ -147,7 +153,7 @@ def get_PDF_examples(directory):
|
|
| 147 |
example_pdf_files = []
|
| 148 |
for filename in os.listdir(directory):
|
| 149 |
if filename.endswith('.pdf'):
|
| 150 |
-
example_pdf_files.append(os.path.join(directory, filename))
|
| 151 |
return example_pdf_files
|
| 152 |
|
| 153 |
|
|
@@ -155,7 +161,7 @@ def get_PDF_examples(directory):
|
|
| 155 |
class FindInvoiceNumberColumns(dspy.Signature):
|
| 156 |
"""Given an input remittance letter, return a list of column header names that may contain invoice numbers."""
|
| 157 |
content = dspy.InputField(desc="remittance letter", format=lambda s:s) # s:s so it doesn't skip the new lines
|
| 158 |
-
column_header_names = dspy.OutputField(desc="comma-separated list of column header names that may contain "
|
| 159 |
"invoice numbers")
|
| 160 |
|
| 161 |
class InvoiceColumnHeaders(dspy.Module):
|
|
@@ -177,7 +183,7 @@ class InvoiceColumnHeaders(dspy.Module):
|
|
| 177 |
return Prediction(column_header_names=unique_headers)
|
| 178 |
|
| 179 |
class FindInvoiceList(dspy.Signature):
|
| 180 |
-
"""Given an input remittance letter and a column header name output a comma-separated list of all invoice numbers """
|
| 181 |
"""that belong to that column."""
|
| 182 |
content = dspy.InputField(desc="remittance letter", format=lambda s:s) # s:s so it doesn't skip the new lines
|
| 183 |
invoice_column_header = dspy.InputField(desc="invoice column header name")
|
|
@@ -214,7 +220,7 @@ class InvoiceList(dspy.Module):
|
|
| 214 |
class FindTotalAmountColumns(dspy.Signature):
|
| 215 |
"""Given an input remittance letter, return a list of column header names that may contain the total payment amount."""
|
| 216 |
content = dspy.InputField(desc="remittance letter", format=lambda s:s) # s:s so it doesn't skip the new lines
|
| 217 |
-
total_column_header_names = dspy.OutputField(desc="comma-separated list of column header names that may contain "
|
| 218 |
"the remittance letter total payment amount")
|
| 219 |
|
| 220 |
class TotalAmountColumnHeaders(dspy.Module):
|
|
@@ -227,7 +233,7 @@ class TotalAmountColumnHeaders(dspy.Module):
|
|
| 227 |
return prediction
|
| 228 |
|
| 229 |
class FindTotalAmount(dspy.Signature):
|
| 230 |
-
"""Given an input remittance letter and a column header name output the total payment amount """
|
| 231 |
"""that belongs to that column."""
|
| 232 |
content = dspy.InputField(desc="remittance letter", format=lambda s:s) # s:s so it doesn't skip the new lines
|
| 233 |
total_amount_column_header = dspy.InputField(desc="total amount header name")
|
|
@@ -256,13 +262,13 @@ class RemittanceLetterTotalAmount(dspy.Module):
|
|
| 256 |
return Prediction(candidate_total_amounts=potential_total_amounts)
|
| 257 |
|
| 258 |
|
| 259 |
-
# Pipeline
|
| 260 |
-
def
|
| 261 |
# Get invoice candidates
|
| 262 |
invoice_list_baseline = InvoiceList()
|
| 263 |
candidate_invoices = invoice_list_baseline(file_content=file_content).candidate_invoice_numbers
|
| 264 |
|
| 265 |
-
candidate_invoices = [
|
| 266 |
|
| 267 |
# Get total amount candidates
|
| 268 |
total_amount_baseline = RemittanceLetterTotalAmount()
|
|
@@ -273,35 +279,94 @@ def poc_production_pipeline_without_verification(file_content):
|
|
| 273 |
# Only keep unique amounts
|
| 274 |
candidate_total_amounts = list(set(candidate_total_amounts))
|
| 275 |
|
| 276 |
-
#
|
| 277 |
-
|
| 278 |
-
|
| 279 |
|
|
|
|
| 280 |
for candidate in candidate_invoices:
|
| 281 |
-
|
|
|
|
| 282 |
|
|
|
|
|
|
|
|
|
|
| 283 |
for candidate in candidate_total_amounts:
|
| 284 |
-
|
| 285 |
-
|
| 286 |
-
return candidate_invoices_for_UI, candidate_total_amounts_for_UI
|
| 287 |
-
|
| 288 |
-
def poc_production_pipeline_without_verification_from_PDF(file_path):
|
| 289 |
-
file_content = extract_text_using_pdfplumber(file_path)
|
| 290 |
-
return poc_production_pipeline_without_verification(file_content)
|
| 291 |
-
|
| 292 |
|
| 293 |
-
#
|
| 294 |
-
|
|
|
|
|
|
|
|
|
|
| 295 |
|
| 296 |
-
|
| 297 |
-
poc_production_pipeline_without_verification_from_PDF,
|
| 298 |
-
[PDF(label="Remittance advice", height=1000)],
|
| 299 |
-
[
|
| 300 |
-
gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Retrieved invoice proposals"], wrap=True),
|
| 301 |
-
gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Retrieved total amount proposals"], wrap=True)
|
| 302 |
-
],
|
| 303 |
-
examples=fake_PDF_examples,
|
| 304 |
-
allow_flagging='never'
|
| 305 |
-
)
|
| 306 |
|
| 307 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
|
| 26 |
pdf_examples_dir = './pdfexamples/'
|
| 27 |
|
| 28 |
+
# model = dspy.LM(
|
| 29 |
+
# model='gpt-3.5-turbo',
|
| 30 |
+
# api_key=os.getenv('OPENAI_PROJECT_KEY'),
|
| 31 |
+
# max_tokens=2000,
|
| 32 |
+
# temperature=0.01)
|
| 33 |
+
|
| 34 |
model = dspy.OpenAI(
|
| 35 |
model='gpt-3.5-turbo-0125',
|
| 36 |
api_key=os.getenv('OPENAI_PROJECT_KEY'),
|
|
|
|
| 153 |
example_pdf_files = []
|
| 154 |
for filename in os.listdir(directory):
|
| 155 |
if filename.endswith('.pdf'):
|
| 156 |
+
example_pdf_files.append([os.path.join(directory, filename), '', ''])
|
| 157 |
return example_pdf_files
|
| 158 |
|
| 159 |
|
|
|
|
| 161 |
class FindInvoiceNumberColumns(dspy.Signature):
|
| 162 |
"""Given an input remittance letter, return a list of column header names that may contain invoice numbers."""
|
| 163 |
content = dspy.InputField(desc="remittance letter", format=lambda s:s) # s:s so it doesn't skip the new lines
|
| 164 |
+
column_header_names = dspy.OutputField(desc="comma-separated list of column header names that may contain "
|
| 165 |
"invoice numbers")
|
| 166 |
|
| 167 |
class InvoiceColumnHeaders(dspy.Module):
|
|
|
|
| 183 |
return Prediction(column_header_names=unique_headers)
|
| 184 |
|
| 185 |
class FindInvoiceList(dspy.Signature):
|
| 186 |
+
"""Given an input remittance letter and a column header name output a comma-separated list of all invoice numbers """
|
| 187 |
"""that belong to that column."""
|
| 188 |
content = dspy.InputField(desc="remittance letter", format=lambda s:s) # s:s so it doesn't skip the new lines
|
| 189 |
invoice_column_header = dspy.InputField(desc="invoice column header name")
|
|
|
|
| 220 |
class FindTotalAmountColumns(dspy.Signature):
|
| 221 |
"""Given an input remittance letter, return a list of column header names that may contain the total payment amount."""
|
| 222 |
content = dspy.InputField(desc="remittance letter", format=lambda s:s) # s:s so it doesn't skip the new lines
|
| 223 |
+
total_column_header_names = dspy.OutputField(desc="comma-separated list of column header names that may contain "
|
| 224 |
"the remittance letter total payment amount")
|
| 225 |
|
| 226 |
class TotalAmountColumnHeaders(dspy.Module):
|
|
|
|
| 233 |
return prediction
|
| 234 |
|
| 235 |
class FindTotalAmount(dspy.Signature):
|
| 236 |
+
"""Given an input remittance letter and a column header name output the total payment amount """
|
| 237 |
"""that belongs to that column."""
|
| 238 |
content = dspy.InputField(desc="remittance letter", format=lambda s:s) # s:s so it doesn't skip the new lines
|
| 239 |
total_amount_column_header = dspy.InputField(desc="total amount header name")
|
|
|
|
| 262 |
return Prediction(candidate_total_amounts=potential_total_amounts)
|
| 263 |
|
| 264 |
|
| 265 |
+
# Pipeline with Verification
|
| 266 |
+
def poc_production_pipeline_with_verification(file_content, verification_invoices, verification_total_amount):
|
| 267 |
# Get invoice candidates
|
| 268 |
invoice_list_baseline = InvoiceList()
|
| 269 |
candidate_invoices = invoice_list_baseline(file_content=file_content).candidate_invoice_numbers
|
| 270 |
|
| 271 |
+
candidate_invoices = [','.join(sorted(lst)) for lst in candidate_invoices]
|
| 272 |
|
| 273 |
# Get total amount candidates
|
| 274 |
total_amount_baseline = RemittanceLetterTotalAmount()
|
|
|
|
| 279 |
# Only keep unique amounts
|
| 280 |
candidate_total_amounts = list(set(candidate_total_amounts))
|
| 281 |
|
| 282 |
+
# Verify invoices
|
| 283 |
+
verification_invoices_list = parse_CSV_string_to_unique(verification_invoices)
|
| 284 |
+
verification_invoices_list_sorted = ','.join(sorted(verification_invoices_list))
|
| 285 |
|
| 286 |
+
validated_invoices = []
|
| 287 |
for candidate in candidate_invoices:
|
| 288 |
+
if candidate == verification_invoices_list_sorted:
|
| 289 |
+
validated_invoices.append(candidate)
|
| 290 |
|
| 291 |
+
# Verify total amount
|
| 292 |
+
verification_total_amount_formatted = format_text_decimal(verification_total_amount)
|
| 293 |
+
validated_total_amount = []
|
| 294 |
for candidate in candidate_total_amounts:
|
| 295 |
+
if candidate == verification_total_amount_formatted:
|
| 296 |
+
validated_total_amount.append(candidate)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 297 |
|
| 298 |
+
# Prepare output for UI
|
| 299 |
+
candidate_invoices_for_UI = [(candidate,) for candidate in candidate_invoices]
|
| 300 |
+
candidate_total_amounts_for_UI = [(candidate,) for candidate in candidate_total_amounts]
|
| 301 |
+
validated_invoices_for_UI = [(validated,) for validated in validated_invoices]
|
| 302 |
+
validated_total_amount_for_UI = [(validated,) for validated in validated_total_amount]
|
| 303 |
|
| 304 |
+
return candidate_invoices_for_UI, candidate_total_amounts_for_UI, validated_invoices_for_UI, validated_total_amount_for_UI
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 305 |
|
| 306 |
+
def poc_production_pipeline_with_verification_from_PDF(file_path, verification_invoices, verification_total_amount):
|
| 307 |
+
file_content = extract_text_using_pdfplumber(file_path)
|
| 308 |
+
return poc_production_pipeline_with_verification(file_content, verification_invoices, verification_total_amount)
|
| 309 |
+
|
| 310 |
+
|
| 311 |
+
|
| 312 |
+
# Main app function
|
| 313 |
+
def main():
|
| 314 |
+
fake_PDF_examples = get_PDF_examples(pdf_examples_dir)
|
| 315 |
+
|
| 316 |
+
# remittance_letter_demo_with_verification_from_PDF = gr.Interface(
|
| 317 |
+
# poc_production_pipeline_with_verification_from_PDF,
|
| 318 |
+
# [
|
| 319 |
+
# PDF(label="Remittance advice", height=800),
|
| 320 |
+
# gr.Textbox(label="Verification Invoices (comma-separated)", placeholder="Enter invoice numbers here..."),
|
| 321 |
+
# gr.Textbox(label="Verification Total Amount", placeholder="Enter total amount here...")
|
| 322 |
+
# ],
|
| 323 |
+
# [
|
| 324 |
+
# gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Retrieved Invoice Proposals"], wrap=True),
|
| 325 |
+
# gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Retrieved Total Amount Proposals"], wrap=True),
|
| 326 |
+
# gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Validated Invoices"], wrap=True),
|
| 327 |
+
# gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Validated Total Amount"], wrap=True)
|
| 328 |
+
# ],
|
| 329 |
+
# examples=fake_PDF_examples,
|
| 330 |
+
# allow_flagging='never'
|
| 331 |
+
# )
|
| 332 |
+
|
| 333 |
+
with gr.Blocks() as remittance_demo:
|
| 334 |
+
gr.Markdown("# Remittance PDF Processor")
|
| 335 |
+
gr.Markdown("Upload a PDF file to extract invoice numbers and payment amounts. Provide verification data if available for comparison.")
|
| 336 |
+
|
| 337 |
+
with gr.Row():
|
| 338 |
+
with gr.Column():
|
| 339 |
+
pdf_input = PDF(label="Remittance advice", height=900)
|
| 340 |
+
|
| 341 |
+
with gr.Column():
|
| 342 |
+
with gr.Accordion("Verification Inputs", open=False):
|
| 343 |
+
verification_invoices = gr.Textbox(label="Verification Invoices (comma-separated)", placeholder="Enter invoice numbers here...")
|
| 344 |
+
verification_total_amount = gr.Textbox(label="Verification Total Amount", placeholder="Enter total amount here...")
|
| 345 |
+
|
| 346 |
+
retrieved_invoices = gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Retrieved Invoice Proposals"], wrap=True)
|
| 347 |
+
retrieved_amounts = gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Retrieved Total Amount Proposals"], wrap=True)
|
| 348 |
+
validated_invoices = gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Validated Invoices"], wrap=True)
|
| 349 |
+
validated_total_amount = gr.Dataframe(col_count=(1, 'fixed'), label="", headers=["Validated Total Amount"], wrap=True)
|
| 350 |
+
|
| 351 |
+
submit_button = gr.Button("Process document")
|
| 352 |
+
|
| 353 |
+
submit_button.click(
|
| 354 |
+
poc_production_pipeline_with_verification_from_PDF,
|
| 355 |
+
inputs=[pdf_input, verification_invoices, verification_total_amount],
|
| 356 |
+
outputs=[retrieved_invoices, retrieved_amounts, validated_invoices, validated_total_amount]
|
| 357 |
+
)
|
| 358 |
+
|
| 359 |
+
gr.Examples(
|
| 360 |
+
examples=[[pdf[0]] for pdf in fake_PDF_examples], # We do this so only PDFs are shown
|
| 361 |
+
inputs=[pdf_input],
|
| 362 |
+
outputs=[retrieved_invoices, retrieved_amounts, validated_invoices, validated_total_amount],
|
| 363 |
+
fn=poc_production_pipeline_with_verification_from_PDF,
|
| 364 |
+
cache_examples=True
|
| 365 |
+
)
|
| 366 |
+
remittance_demo.launch()
|
| 367 |
+
|
| 368 |
+
|
| 369 |
+
|
| 370 |
+
# Run the main app if the file is executed directly
|
| 371 |
+
if __name__ == "__main__":
|
| 372 |
+
main()
|