main.py

from fastapi import FastAPI, Request, Header, HTTPException, Query
from fastapi.responses import HTMLResponse, JSONResponse
from fastapi.openapi.docs import get_swagger_ui_html
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from datetime import datetime
import uuid
import os
import openai
 
from transformers import pipeline
import logging, traceback
from typing import Optional, List, Union

from model import (
    summarize_review, smart_summarize, detect_industry,
    detect_product_category, detect_emotion, answer_followup, answer_only,
    assess_churn_risk, extract_pain_points  # ✅ Added extract_pain_points
)

app = FastAPI(
    title="🧠 ChurnSight AI",
    description="Multilingual GenAI for smarter feedback — summarization, sentiment, emotion, aspects, Q&A and tags.",
    version="2025.1.0",
    openapi_url="/openapi.json",
    docs_url=None,
    redoc_url="/redoc"
)

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

logging.basicConfig(level=logging.INFO)
VALID_API_KEY = "my-secret-key"
log_store = []  # ✅ Shared in-memory churn log

@app.get("/", response_class=HTMLResponse)
def root():
    return "<h1>ChurnSight AI Backend is Running</h1>"

@app.get("/docs", include_in_schema=False)
def custom_swagger_ui():
    return get_swagger_ui_html(
        openapi_url=app.openapi_url,
        title="🧠 Swagger UI - ChurnSight AI",
        swagger_favicon_url="https://cdn-icons-png.flaticon.com/512/3794/3794616.png",
        swagger_js_url="https://cdn.jsdelivr.net/npm/[email protected]/swagger-ui-bundle.js",
        swagger_css_url="https://cdn.jsdelivr.net/npm/[email protected]/swagger-ui.css",
    )

@app.exception_handler(Exception)
async def exception_handler(request: Request, exc: Exception):
    logging.error(f"Unhandled Exception: {traceback.format_exc()}")
    return JSONResponse(status_code=500, content={"detail": "Internal Server Error. Please contact support."})

# ==== SCHEMAS ====

class ReviewInput(BaseModel):
    text: str
    model: str = "distilbert-base-uncased-finetuned-sst-2-english"
    industry: Optional[str] = None
    aspects: bool = False
    follow_up: Optional[Union[str, List[str]]] = None
    product_category: Optional[str] = None
    device: Optional[str] = None
    intelligence: Optional[bool] = False
    verbosity: Optional[str] = "detailed"

class BulkReviewInput(BaseModel):
    reviews: List[str]
    model: str = "distilbert-base-uncased-finetuned-sst-2-english"
    industry: Optional[List[str]] = None
    aspects: bool = False
    product_category: Optional[List[str]] = None
    device: Optional[List[str]] = None
    follow_up: Optional[List[Union[str, List[str]]]] = None
    intelligence: Optional[bool] = False
    explain_bulk: Optional[bool] = False

class FollowUpRequest(BaseModel):
    text: str
    question: str
    verbosity: Optional[str] = "brief"

# ==== HELPERS ====

def auto_fill(value: Optional[str], fallback: str) -> str:
    if not value or value.lower() == "auto-detect":
        return fallback
    return value

# ==== ENDPOINTS ====

@app.post("/analyze/")
async def analyze(data: ReviewInput, x_api_key: str = Header(None)):
    if x_api_key and x_api_key != VALID_API_KEY:
        raise HTTPException(status_code=401, detail="❌ Invalid API key")

    if len(data.text.split()) < 20:
        raise HTTPException(status_code=400, detail="⚠️ Review too short for analysis (min. 20 words).")

    global log_store

    try:
        # === Generate Summary ===
        summary = (
            summarize_review(data.text, max_len=40, min_len=8)
            if data.verbosity.lower() == "brief"
            else smart_summarize(data.text, n_clusters=2 if data.intelligence else 1)
        )

        # === Sentiment + Emotion ===
        sentiment_pipeline = pipeline("sentiment-analysis", model=data.model)
        sentiment = sentiment_pipeline(data.text)[0]

        emotion = detect_emotion(data.text)


        churn_risk = assess_churn_risk(sentiment["label"], emotion)

        # === Auto-detect metadata ===
        industry = detect_industry(data.text) if not data.industry or "auto" in data.industry.lower() else data.industry
        product_category = detect_product_category(data.text) if not data.product_category or "auto" in data.product_category.lower() else data.product_category

        # === Optional: Pain Points ===
        pain_points = extract_pain_points(data.text) if data.aspects else []

        # === Log entry ===
        log_store.append({
            "timestamp": datetime.now(),
            "product": product_category,
            "churn_risk": churn_risk,
            "user_id": str(uuid.uuid4())
        })
        if len(log_store) > 1000:
            log_store = log_store[-1000:]

        # === Final API Response ===
        response = {
            "summary": summary,
            "sentiment": sentiment,
            "emotion": emotion,
            "product_category": product_category,
            "device": "Web",
            "industry": industry,
            "churn_risk": churn_risk,
            "pain_points": pain_points
        }

        if data.follow_up:
            response["follow_up"] = answer_followup(data.text, data.follow_up, verbosity=data.verbosity)

        return response

    except Exception as e:
        logging.error(f"🔥 Unexpected analysis failure: {traceback.format_exc()}")
        raise HTTPException(status_code=500, detail="Internal Server Error during analysis.")

@app.post("/followup/")
async def followup(request: FollowUpRequest, x_api_key: str = Header(None)):
    if x_api_key and x_api_key != VALID_API_KEY:
        raise HTTPException(status_code=401, detail="Invalid API key")
    try:
        if not request.question or len(request.text.split()) < 10:
            raise HTTPException(status_code=400, detail="Question or text is too short.")
        return {"answer": answer_only(request.text, request.question)}
    except Exception as e:
        logging.error(f"❌ Follow-up failed: {traceback.format_exc()}")
        raise HTTPException(status_code=500, detail="Follow-up generation failed.")

@app.get("/log/")
async def get_churn_log(x_api_key: str = Header(None)):
    if x_api_key and x_api_key != VALID_API_KEY:
        raise HTTPException(status_code=401, detail="Unauthorized")
    return {"log": log_store}

@app.post("/bulk/")
async def bulk_analyze(data: BulkReviewInput, token: str = Query(None)):
    if token != VALID_API_KEY:
        raise HTTPException(status_code=401, detail="❌ Unauthorized: Invalid API token")

    global log_store

    try:
        results = []
        sentiment_pipeline = pipeline("sentiment-analysis", model=data.model)

        for i, review_text in enumerate(data.reviews):
            if not review_text.strip():
                continue  # Skip empty reviews

            if len(review_text.split()) < 20:
                results.append({
                    "review": review_text,
                    "error": "Too short to analyze"
                })
                continue


            summary = smart_summarize(review_text, n_clusters=2 if data.intelligence else 1)
            sentiment = sentiment_pipeline(review_text)[0]

            emotion = detect_emotion(review_text)


            churn = assess_churn_risk(sentiment["label"], emotion)
            pain = extract_pain_points(review_text) if data.aspects else []

            ind = auto_fill(data.industry[i] if data.industry else None, detect_industry(review_text))
            prod = auto_fill(data.product_category[i] if data.product_category else None, detect_product_category(review_text))
            dev = auto_fill(data.device[i] if data.device else None, "Web")

            result = {
                "review": review_text,
                "summary": summary,
                "sentiment": sentiment["label"],
                "score": sentiment["score"],
                "emotion": emotion,
                "industry": ind,
                "product_category": prod,
                "device": dev,
                "churn_risk": churn,
                "pain_points": pain
            }

            # ✅ Optional follow-up
            if data.follow_up and i < len(data.follow_up):
                follow_q = data.follow_up[i]
                result["follow_up"] = answer_followup(review_text, follow_q)

            # ✅ Log churn entry
            log_store.append({
                "timestamp": datetime.now(),
                "product": prod,
                "churn_risk": churn,
                "user_id": str(uuid.uuid4())
            })

            results.append(result)

        # ✅ Cap log size
        if len(log_store) > 1000:
            log_store = log_store[-1000:]

        return {"results": results}

    except Exception as e:
        logging.error(f"🔥 Bulk processing failed: {traceback.format_exc()}")
        raise HTTPException(status_code=500, detail="Failed to analyze bulk reviews")

# Already set with os.environ — nothing else needed
@app.post("/rootcause/")
async def root_cause_analysis(payload: dict, x_api_key: str = Header(None)):
    if x_api_key and x_api_key != VALID_API_KEY:
        raise HTTPException(status_code=401, detail="Invalid API key")

    try:
        text = payload.get("text", "").strip()
        if not text or len(text.split()) < 5:
            raise HTTPException(status_code=400, detail="Insufficient input for root cause analysis.")

        prompt = f"""
        Analyze the following customer feedback and extract:
        1. The main problem
        2. The possible root cause
        3. A suggested fix or which team might need to handle it

        Feedback: '''{text}'''
        Format your answer as:
        Problem: ...
        Cause: ...
        Suggestion: ...
        """

        # Models to try in order
        models_to_try = ["gpt-4", "gpt-4o-mini", "gpt-3.5-turbo"]

        last_error = None
        for model_name in models_to_try:
            try:
                response = openai.chat.completions.create(
                    model=model_name,
                    messages=[{"role": "user", "content": prompt}]
                )
                output = response.choices[0].message.content
                lines = output.splitlines()

                def extract_line(tag):
                    for line in lines:
                        if line.lower().startswith(tag.lower()):
                            return line.split(":", 1)[-1].strip()
                    return "—"

                return {
                    "problem": extract_line("Problem"),
                    "cause": extract_line("Cause"),
                    "suggestion": extract_line("Suggestion"),
                    "model_used": model_name
                }
            except Exception as e:
                last_error = str(e)
                logging.warning(f"Model {model_name} failed: {last_error}")
                continue

        # If all models fail
        raise HTTPException(status_code=500, detail=f"All model attempts failed. Last error: {last_error}")

    except Exception as e:
        logging.error(f"Root cause analysis failed: {traceback.format_exc()}")
        return JSONResponse(status_code=500, content={"detail": f"Root cause generation failed: {str(e)}"})