app.py

import numpy as np
from sklearn.ensemble import IsolationForest
import joblib
import gradio as gr

# =====================================================================
# 1. CORE ARCHITECTURE SPECIFICATION: ZEWPOLAI EXC1
# =====================================================================
class ZewpolAI_EXC1_Core:
    def __init__(self, contamination=0.04):
        """
        Instantiates the raw mathematical brain for ZewpolAI EXC1.
        Features tracked: [Vector_0: Request Speed (req/s), Vector_1: Payload Weight (KB)]
        """
        self.model = IsolationForest(
            contamination=contamination, 
            random_state=42, 
            n_estimators=100
        )
        self.is_trained = False

    def compile_and_train(self):
        """Generates the multi-variable cluster array and trains the brain."""
        # Synthesize 1,000 matrix logs representing verified, safe operations
        np.random.seed(42)
        normal_speeds = np.random.normal(loc=2.0, scale=0.8, size=(1000, 1))
        normal_payloads = np.random.normal(loc=150.0, scale=45.0, size=(1000, 1))
        training_matrix = np.hstack((normal_speeds, normal_payloads))
        
        # Train the Isolation Forest to map boundary thresholds
        self.model.fit(training_matrix)
        self.is_trained = True
        
        # Serialize and save model states locally
        joblib.dump(self.model, "zewpol_exc1.pkl")
        return "[SUCCESS] ZewpolAI EXC1 Brain compiled. Boundaries mapped successfully."

    def evaluate_telemetry(self, speed, payload):
        """Runs inference to decide if inbound vectors are a Pass or a Block."""
        if not self.is_trained:
            return "UNTRAINED_ENGINE", "Model must be compiled first."
            
        # Format user inputs into a 2D numpy feature vector
        test_vector = np.array([[float(speed), float(payload)]])
        
        # Inference prediction: 1 = Normal/Safe User, -1 = Rogue Outlier Anomaly
        prediction = self.model.predict(test_vector)
        
        if prediction == 1:
            return "🟢 ALLOW_PASS", "Telemetry matches secure historical baseline clusters."
        else:
            return "🚨 BLOCK_DROP", "Statistical anomaly detected! High-velocity or abnormal data exfiltration risk."

# =====================================================================
# 2. MODEL COMPILATION ON STARTUP
# =====================================================================
# Initialize and train the model immediately when Hugging Face spins up the container
zewpol_brain = ZewpolAI_EXC1_Core()
startup_log = zewpol_brain.compile_and_train()

# =====================================================================
# 3. HUGGING FACE WEB API INTERFACE
# =====================================================================
# We use Gradio to build a clean API interface for our Python model functions
def raw_model_inference_pipeline(req_speed, file_size_kb):
    decision, logs = zewpol_brain.evaluate_telemetry(req_speed, file_size_kb)
    return {
        "Model Compilation Status": startup_log,
        "Firewall Action": decision,
        "Engine Diagnostics": logs
    }

# Build the pipeline UI block
demo = gr.Interface(
    fn=raw_model_inference_pipeline,
    inputs=[
        gr.Number(value=2.5, label="Input Parameter A: Speed (Requests per second)"),
        gr.Number(value=120.0, label="Input Parameter B: Payload Size (KB)")
    ],
    outputs=gr.JSON(label="ZewpolAI EXC1 Real-Time JSON Output Matrix"),
    title="🛡️ ZewpolAI EXC1 Core Model Pipeline",
    description="The raw machine learning engine for ZewpolAI EXC1. Enter web metrics to see the model classify traffic vectors in real-time."
)

if __name__ == "__main__":
    demo.launch()