src/interface/api.py

from fastapi import FastAPI, HTTPException, BackgroundTasks
from pydantic import BaseModel, HttpUrl
from typing import Optional, Dict, List, Any
import logging
import time

from src.core.pipeline import Pipeline
from src.ml import slm_explainer

# Configure Logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger("api")

# Initialize App
app = FastAPI(
    title="PhishingInsight API",
    description="AI-powered Phishing Detection & URL Analysis API",
    version="1.0.0"
)

# Initialize Pipeline (Global Singleton)
logger.info("API: Initializing Global Pipeline...")
pipeline = Pipeline()

# Load SLM in background on startup
@app.on_event("startup")
async def startup_event():
    logger.info("API: Pre-loading SLM Model...")
    # We do this asynchronously or threading to not block, 
    # but slm_explainer.load_model() blocking is safer for "ready" state.
    # For now, we rely on the main process having loaded it previously 
    # or it loading on first request if isolated.
    try:
        slm_explainer.load_model()
        logger.info("API: SLM Model Loaded.")
    except Exception as e:
        logger.error(f"API: Model load warning: {e}")

# --- Data Models ---

class URLRequest(BaseModel):
    url: str

class SignalScores(BaseModel):
    url_analysis: float
    domain_check: float

class AnalysisResponse(BaseModel):
    success: bool
    url: str
    verdict: str  # SAFE, WARNING, DANGER
    score: float  # 0.0 to 1.0 (Risk)
    is_phishing: bool
    
    summary: str
    explanation: str
    advice: str
    
    signals: SignalScores
    findings: List[str]
    
    timestamp: str
    time_ms: int
    
    error: Optional[str] = None

# --- Endpoints ---

@app.get("/health")
def health_check():
    """Health check endpoint for Docker/Kubernetes."""
    return {"status": "healthy", "service": "PhishingInsight-API"}

@app.get("/version")
def version_info():
    """Return API and Model versions."""
    return {
        "api_version": "1.0.0",
        "model_type": "Hybrid (LightGBM + TinyLlama)"
    }

@app.post("/analyze", response_model=AnalysisResponse)
def analyze_url(request: URLRequest):
    """
    Analyze a URL for phishing threats.
    """
    logger.info(f"API Request: Analyze {request.url}")
    
    result = pipeline.analyze(request.url)
    
    if not result.get('success'):
        logger.warning(f"API Analysis Failed: {result.get('error')}")
        # Even if logical failure (invalid URL), we return a structured response with error
        return {
            "success": False,
            "url": request.url,
            "verdict": "ERROR",
            "score": 0.0,
            "is_phishing": False,
            "summary": "Analysis Failed",
            "explanation": result.get('error', "Unknown error"),
            "advice": "Check the URL and try again.",
            "signals": {"url_analysis": 0, "domain_check": 0},
            "findings": [],
            "timestamp": "",
            "time_ms": 0,
            "error": result.get('error')
        }
        
    return result