Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import fitz # PyMuPDF
|
| 2 |
+
import os
|
| 3 |
+
import json
|
| 4 |
+
import gradio as gr
|
| 5 |
+
from groq import Groq
|
| 6 |
+
|
| 7 |
+
# π Initialize Groq Client
|
| 8 |
+
client = Groq(api_key=os.getenv("GROQ_API_KEY"))
|
| 9 |
+
|
| 10 |
+
def extract_form_fields(pdf_bytes):
|
| 11 |
+
doc = fitz.open(stream=pdf_bytes, filetype="pdf")
|
| 12 |
+
form_fields = {}
|
| 13 |
+
for page in doc:
|
| 14 |
+
widgets = page.widgets()
|
| 15 |
+
if widgets:
|
| 16 |
+
for widget in widgets:
|
| 17 |
+
key = widget.field_name
|
| 18 |
+
value = widget.field_value if widget.field_value else ""
|
| 19 |
+
form_fields[key] = value
|
| 20 |
+
return form_fields
|
| 21 |
+
|
| 22 |
+
def get_pdf_text(pdf_bytes):
|
| 23 |
+
doc = fitz.open(stream=pdf_bytes, filetype="pdf")
|
| 24 |
+
return "\n".join([page.get_text() for page in doc])
|
| 25 |
+
|
| 26 |
+
def get_field_details(form_fields, pdf_text):
|
| 27 |
+
prompt = f"""
|
| 28 |
+
You are an expert at analyzing and auto-filling PDF form fields.
|
| 29 |
+
Here is the extracted PDF text:
|
| 30 |
+
{pdf_text}
|
| 31 |
+
|
| 32 |
+
Based on this, explain the meaning or expected value of each of the following fields in JSON format:
|
| 33 |
+
|
| 34 |
+
{json.dumps(list(form_fields.keys()), indent=2)}
|
| 35 |
+
|
| 36 |
+
Return your output in the following JSON format:
|
| 37 |
+
{{ "field_name_1": "description", "field_name_2": "description", ... }}
|
| 38 |
+
"""
|
| 39 |
+
response = client.chat.completions.create(
|
| 40 |
+
model="llama3-8b-8192",
|
| 41 |
+
messages=[{"role": "user", "content": prompt}]
|
| 42 |
+
)
|
| 43 |
+
|
| 44 |
+
return response.choices[0].message.content
|
| 45 |
+
|
| 46 |
+
def analyze_form(pdf_file):
|
| 47 |
+
if not pdf_file:
|
| 48 |
+
return "β No file provided.", "", ""
|
| 49 |
+
|
| 50 |
+
try:
|
| 51 |
+
pdf_bytes = pdf_file.read()
|
| 52 |
+
fields = extract_form_fields(pdf_bytes)
|
| 53 |
+
pdf_text = get_pdf_text(pdf_bytes)
|
| 54 |
+
explanation = get_field_details(fields, pdf_text)
|
| 55 |
+
|
| 56 |
+
return "β
Analysis complete.", json.dumps(fields, indent=2), explanation
|
| 57 |
+
except Exception as e:
|
| 58 |
+
return f"β Error: {str(e)}", "", ""
|
| 59 |
+
|
| 60 |
+
# π¨ Gradio Interface
|
| 61 |
+
with gr.Blocks() as demo:
|
| 62 |
+
gr.Markdown("## π Form Field Analyzer")
|
| 63 |
+
gr.Markdown("Upload a **tax or registration PDF form**. This tool extracts fillable fields and describes what each might represent.")
|
| 64 |
+
|
| 65 |
+
with gr.Row():
|
| 66 |
+
file_input = gr.File(label="Upload PDF", file_types=[".pdf"])
|
| 67 |
+
analyze_btn = gr.Button("Analyze Form")
|
| 68 |
+
|
| 69 |
+
status_output = gr.Textbox(label="Status")
|
| 70 |
+
fields_output = gr.Code(label="π Extracted Fields (JSON)", language="json")
|
| 71 |
+
desc_output = gr.Code(label="π‘ Field Descriptions (LLM Response)", language="json")
|
| 72 |
+
|
| 73 |
+
analyze_btn.click(fn=analyze_form, inputs=[file_input], outputs=[status_output, fields_output, desc_output])
|
| 74 |
+
|
| 75 |
+
demo.launch()
|