app_deepseek.py

import gradio as gr
import openpyxl
import PyPDF2
import pandas as pd
from PIL import Image
import pytesseract  # Replaced EasyOCR
import io
import os
from huggingface_hub import InferenceClient

# Access the Hugging Face token from the environment variable
hf_token = os.environ.get("HF_TOKEN")


def extract_transactions(file):
    """Extract raw text content from various supported file formats."""
    filename = file.name.lower()

    try:
        if filename.endswith((".xlsx", ".xls")):
            workbook = openpyxl.load_workbook(file, data_only=True)
            sheet = workbook.active
            return "\n".join(
                "\t".join(str(cell.value or "") for cell in row)
                for row in sheet.iter_rows()
            )

        elif filename.endswith(".csv"):
            df = pd.read_csv(file)
            return df.to_string(index=False)

        elif filename.endswith(".pdf"):
            reader = PyPDF2.PdfReader(file)
            return "\n".join(page.extract_text() or "" for page in reader.pages)

        elif filename.endswith((".jpg", ".jpeg", ".png")):
            image = Image.open(io.BytesIO(file.read()))
            return pytesseract.image_to_string(image)

        else:
            raise ValueError("Unsupported file format.")

    except Exception as e:
        raise ValueError(f"Error processing {file.name}: {e}")


def reconcile_statements(erp_file, bank_file):
    yield "⏳ Processing your request... Please wait.", ""

    # your existing code block...

    try:
        erp_statement = extract_transactions(erp_file)
        bank_statement = extract_transactions(bank_file)
        print("ERP statement is: ")
        print(erp_statement)

        print("Bank statement is: ")
        print(bank_statement)

        # Hugging Face request...
        #prompt = f"Reconcile these statements:\nERP:\n{erp_statement}\nBank:\n{bank_statement}"
        prompt = f"""You are a financial analyst specializing in account reconciliations. Your task is to compare two data sets: one from an ERP system and the other from a bank statement.

The goal is to identify which transactions match across both data sets, and which transactions are unmatched or potentially erroneous.

Each dataset contains transaction entries with **Date**, **Amount**, and **Description**.

ERP descriptions may include prefixes like "Vendor Payment - ", while the bank descriptions are simpler (e.g., "Utilities" instead of "Vendor Payment - Utilities"). Please normalize these for accurate comparison.

---

Please follow this format in your response:

1. 📘 **Introduction**  
   Briefly explain what reconciliation means and how you'll approach it.

2. 📄 **Unmatched ERP Transactions**  
   List any transactions found in the ERP file but not in the Bank file.

3. 🏦 **Unmatched Bank Transactions**  
   List any transactions found in the Bank file but not in the ERP file.

4. ✅ **Matched Transactions (optional, summary only)**  
   How many matched successfully?

5. 🧾 **Summary & Suggested Next Steps**  
   Explain what the discrepancies might mean and what the user should do next.

---

Here is the ERP data:
{erp_statement}

---

Here is the Bank data:
{bank_statement}

---

Generate a clean and professional reconciliation report, structured as described above. Make sure the response is easy to read."""


        client = InferenceClient(provider="together", api_key=hf_token)
        completion = client.chat.completions.create(
            model="deepseek-ai/DeepSeek-R1",
            messages=[{"role": "user", "content": prompt}],
            max_tokens=100000
        )

        if completion.choices:
            reconciliation_results = completion.choices[0].message.get('content', '')
            print(reconciliation_results)
        else:
            reconciliation_results = "⚠️ No response received from the model."

        output = f"""
        <div style="font-family: 'Segoe UI', ...">
            <h2>🔍 Reconciliation Results</h2>
            <div style="...">
                <pre>{reconciliation_results}</pre>
            </div>
        </div>
        """
        yield "✅ Processing complete!", output

    except Exception as e:
        yield f"❌ Error: {e}", f"<h1>Error</h1><p>{e}</p>"

with gr.Blocks(css="""
#company-logo {
    width: 25%;
    margin: auto;
    display: block;
}
""") as iface:
    gr.Image("logo_Icon.png", elem_id="company-logo", label="Beiing Human")
    status_text = gr.Markdown("👋 Upload your files to begin reconciliation.")
    with gr.Row():
        erp_input = gr.File(label="📂 Upload ERP Statement", type="filepath")
        bank_input = gr.File(label="📂 Upload Bank Statement", type="filepath")
    submit_btn = gr.Button("🔄 Start Reconciliation")
    result_output = gr.HTML()

    submit_btn.click(
        fn=reconcile_statements,
        inputs=[erp_input, bank_input],
        outputs=[status_text, result_output]
    )

if __name__ == "__main__":
    iface.launch(debug=True, share=True)