Backend/main.py

import os
import torch
from fastapi import FastAPI
from pydantic import BaseModel
from fastapi import UploadFile, File
from fastapi.middleware.cors import CORSMiddleware
from fastapi import UploadFile, File
from inference import extract_text_from_pdf, split_text_by_language, predict_idiom, normalize_text, load_model, IdiomMatcher
from nltk.tokenize import sent_tokenize
from langdetect import detect
from fastapi import HTTPException 
import re
import fitz  # PyMuPDF
# Allow requests from your React app
origins = [
    "http://localhost:3000",  # React dev server
    "https://language-learning-base-website.vercel.app",
    "https://www.idiomator.com"
    "https://idiomator.com"

]

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins= origins,  # or ["*"] for all origins (not recommended in production)
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# Load model once at startup
device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
checkpoint_path = os.path.join(os.path.dirname(__file__), "checkpoints")
model, tokenizer = load_model(checkpoint_path)
model = model.to(device)
model.eval()

class TextRequest(BaseModel):
    text: str
    language: str = "en"  # Default to English

class IdiomResponse(BaseModel):
    idioms: list[str]
    language: str = "en"  # Default to English

@app.get("/")
def root():
    return {"status": "ok"}


@app.post("/extract_idioms_ai", response_model=IdiomResponse)
def extract_idioms(request: TextRequest):
    import time
    start = time.time()
    print(f"[📥] Request received at: {start}")

    text = normalize_text(request.text)
    language = request.language.lower()  # Get the user-selected language

    sentences = split_text_by_language(text, language=language)
    idioms = []
    for sent in sentences:
        idioms.extend(predict_idiom(sent, model, tokenizer, device))
    print(f"[✅] Done in {time.time() - start:.3f}s")
    return {"idioms": idioms}


from fastapi import Form

def check_pdf_page_limit(pdf_bytes, max_pages=10):
    with fitz.open(stream=pdf_bytes, filetype="pdf") as doc:
        if len(doc) > max_pages:
            raise HTTPException(status_code=400, detail=f"PDF has {len(doc)} pages. Limit is {max_pages}.")

@app.post("/extract_idioms_pdf_ai", response_model=IdiomResponse)
async def extract_idioms_pdf(
    file: UploadFile = File(...),
    language: str = Form(...)  # ✅ Get language from the client
):
    
    pdf_bytes = await file.read()
    check_pdf_page_limit(pdf_bytes, max_pages=10)
    text = extract_text_from_pdf(pdf_bytes)
    # Normalize the extracted text!
    text = normalize_text(text)
    sentences = split_text_by_language(text, language=language)
    idioms = []
    for sent in sentences:
        idioms.extend(predict_idiom(sent, model, tokenizer, device))
    return {"idioms": idioms}


idiom_matcher = IdiomMatcher({
    "en": "idioms_structured_1/seed_idioms_en_cleaned.jsonl",
    "es": "idioms_structured_1/seed_idioms_es_cleaned.jsonl"
})

@app.post("/extract_idioms_heuristic", response_model=IdiomResponse)
def extract_idioms_heuristic(request: TextRequest):
    text = normalize_text(request.text)
    language = request.language.lower()  # get the language from request
    idiom_matches = idiom_matcher.match(text, lang=language)
    idioms = [idiom["idiom"] for idiom in idiom_matches]

    return {"idioms": idioms}
@app.post("/extract_idioms_pdf_heuristic", response_model=IdiomResponse)
async def extract_idioms_pdf_(
    file: UploadFile = File(...),
    language: str = Form(...)  # ✅ Get language from the client
):
    pdf_bytes = await file.read()
    check_pdf_page_limit(pdf_bytes, max_pages=10)
    text = extract_text_from_pdf(pdf_bytes)
    # Normalize the extracted text!
    text = normalize_text(text)
    sentences = split_text_by_language(text, language=language)
    idioms = []
    idiom_matches = idiom_matcher.match(text, lang=language) 
    idioms = [idiom["idiom"] for idiom in idiom_matches]
    return {"idioms": idioms}