app.py

import os
import json
from transformers import BertTokenizer, BertForSequenceClassification
import torch
from flask import Flask, request, jsonify
from pathlib import Path
from datetime import datetime

# Initialize Flask app
app = Flask(__name__)

# Load pre-trained model and tokenizer
MODEL_PATH = "path/to/your/model"  # Update with your Hugging Face model path
tokenizer = BertTokenizer.from_pretrained(MODEL_PATH)
model = BertForSequenceClassification.from_pretrained(MODEL_PATH)

# Function to process contract text and classify clauses
def classify_clause(contract_text):
    inputs = tokenizer(contract_text, return_tensors="pt", truncation=True, padding=True, max_length=512)
    with torch.no_grad():
        outputs = model(**inputs)
    logits = outputs.logits
    predicted_class = torch.argmax(logits, dim=-1).item()  # Get predicted class (risk tag)
    
    # Define risk labels (assuming 3 risk levels: low, medium, high)
    risk_labels = ["low", "medium", "high"]
    predicted_risk = risk_labels[predicted_class]

    # Get confidence score (softmax output)
    softmax = torch.nn.Softmax(dim=-1)
    confidence = softmax(logits).squeeze().tolist()[predicted_class]

    return {"predicted_risk": predicted_risk, "confidence_score": confidence}

# Define route to handle file uploads
@app.route("/upload_contract", methods=["POST"])
def upload_contract():
    # Extract file from the request
    if 'file' not in request.files:
        return jsonify({"error": "No file part"}), 400
    
    file = request.files['file']
    
    if file.filename == '':
        return jsonify({"error": "No selected file"}), 400
    
    contract_text = file.read().decode('utf-8')  # Assuming the file is a text-based contract
    
    # Classify the contract text
    result = classify_clause(contract_text)

    # Prepare JSON response
    response_data = {
        "contract_title": "Sample Contract",  # Placeholder, can be parsed from the file
        "overall_risk_score": result["predicted_risk"],  # Risk classification
        "high_risk_clauses": ["Termination Clause", "Penalty Clause"],  # Example (this should be dynamically extracted)
        "risk_map_url": "https://example.com/risk_map",  # Placeholder (use actual URL for visualization)
        "evaluation_date": datetime.now().strftime("%Y-%m-%d")
    }

    # Return response as JSON
    return jsonify(response_data)

if __name__ == "__main__":
    app.run(debug=True, host="0.0.0.0", port=5000)