main.py

import os
from fastapi import FastAPI
from pydantic import BaseModel
from transformers import pipeline, AutoTokenizer
from langdetect import detect, DetectorFactory

# Ensure consistent language detection results
DetectorFactory.seed = 0

# Set Hugging Face cache directory to a writable location
os.environ["HF_HOME"] = "/tmp/huggingface"
os.environ["TRANSFORMERS_CACHE"] = "/tmp/huggingface"
os.makedirs(os.environ["HF_HOME"], exist_ok=True)

app = FastAPI()

# Model names
multilingual_model_name = "johndoee/sentiment"
english_model_name = "siebert/sentiment-roberta-large-english"

# Load tokenizer and model for multilingual sentiment analysis
multilingual_tokenizer = AutoTokenizer.from_pretrained(multilingual_model_name)
multilingual_model = pipeline(
    "sentiment-analysis",
    model=multilingual_model_name,
    tokenizer=multilingual_tokenizer
)

# Load English sentiment model
english_model = pipeline("sentiment-analysis", model=english_model_name)

class SentimentRequest(BaseModel):
    text: str

class SentimentResponse(BaseModel):
    original_text: str
    language_detected: str
    sentiment: str
    confidence_score: float

def detect_language(text):
    try:
        return detect(text)
    except Exception:
        return "unknown"

@app.get("/")
def home():
    return {"message": "Sentiment Analysis API is running!"}

@app.post("/analyze/", response_model=SentimentResponse)
def analyze_sentiment(request: SentimentRequest):
    text = request.text
    language = detect_language(text)

    # Choose the appropriate model based on detected language
    model = english_model if language == "en" else multilingual_model
    result = model(text)

    return SentimentResponse(
        original_text=text,
        language_detected=language,
        sentiment=result[0]["label"].lower(),
        confidence_score=result[0]["score"],
    )