app.py

from flask import Flask, request, jsonify
from flask_cors import CORS
import os
import re
import pandas as pd
from typing import List
from deep_translator import GoogleTranslator

from langchain_core.documents import Document
from langchain_community.document_loaders import (
    WebBaseLoader, PyPDFLoader, Docx2txtLoader
)
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import Chroma
from langchain.chains import RetrievalQA
from langchain.prompts import PromptTemplate
from langchain.llms import HuggingFacePipeline
from transformers import pipeline

from huggingface_hub import snapshot_download

# ============================================================
# APP INIT
# ============================================================

app = Flask(__name__)
CORS(app)

# ============================================================
# CONSTANTS
# ============================================================

DATASET_REPO = "bshk57/Sastra_data"
LOCAL_DATASET_DIR = "knowledge_base"
VECTOR_DB_PATH = "sastra_vector_db"

EMBEDDING_MODEL = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
LLM_MODEL = "google/flan-t5-large"

MANDATORY_URLS = [
    "https://www.sastra.edu/admissions/ug-pg.html",
    "https://www.sastra.edu/admissions/eligibility-criteria.html",
    "https://www.sastra.edu/admissions/fee-structure.html",
]

SASTRA_URLS = [
    "https://www.sastra.edu/about-us.html",
    "https://www.sastra.edu/academics/schools.html#school-of-computing",
    "https://www.sastra.edu/admissions/ug-pg.html",
    "https://www.sastra.edu/admissions/eligibility-criteria.html",
    "https://www.sastra.edu/admissions/fee-structure.html",
    "https://www.sastra.edu/admissions/hostel-fees.html",
    "https://www.sastra.edu/infrastructure/physical-facilities.html",
    "https://www.sastra.edu/about-us/mission-vision.html",
]

os.makedirs(LOCAL_DATASET_DIR, exist_ok=True)

# ============================================================
# UTILITIES
# ============================================================

def extract_urls(text: str):
    return re.findall(r'https?://[^\s]+', text)


def clean_llm_output(text: str):
    text = re.sub(r'^(Answer:|Response:)', '', text, flags=re.I).strip()
    if text.lower().startswith("insufficient_data"):
        return ""
    return re.sub(r'\s+', ' ', text)[:700]


# ============================================================
# LOAD DATASET FROM HUGGINGFACE
# ============================================================

print("⬇ Downloading dataset...")
snapshot_download(
    repo_id=DATASET_REPO,
    repo_type="dataset",
    local_dir=LOCAL_DATASET_DIR,
    local_dir_use_symlinks=False,
    ignore_patterns=[".gitattributes"]
)

# ============================================================
# LOAD LOCAL DOCUMENTS
# ============================================================

def load_local_documents():
    docs = []

    for file in os.listdir(LOCAL_DATASET_DIR):
        path = os.path.join(LOCAL_DATASET_DIR, file)
        try:
            if file.lower().endswith(".pdf"):
                docs.extend(PyPDFLoader(path).load())

            elif file.lower().endswith(".docx"):
                docs.extend(Docx2txtLoader(path).load())

            elif file.lower().endswith(".xlsx"):
                df = pd.read_excel(path)
                for _, row in df.iterrows():
                    text = " | ".join(
                        f"{col}: {row[col]}"
                        for col in df.columns
                        if pd.notna(row[col])
                    )
                    docs.append(Document(
                        page_content=text,
                        metadata={"source": file}
                    ))
        except Exception as e:
            print(f"⚠ Error loading {file}: {e}")

    return docs


# ============================================================
# INITIALIZE RAG MODEL
# ============================================================

def initialize_model():
    global vectordb, qa_chain

    documents = []

    # Website data
    for url in SASTRA_URLS:
        try:
            documents.extend(WebBaseLoader(url).load())
        except:
            pass

    # Dataset + uploaded docs
    documents.extend(load_local_documents())

    splitter = RecursiveCharacterTextSplitter(
        chunk_size=600,
        chunk_overlap=50
    )
    chunks = splitter.split_documents(documents)

    embeddings = HuggingFaceEmbeddings(
        model_name=EMBEDDING_MODEL
    )

    vectordb = Chroma.from_documents(
        chunks,
        embeddings,
        persist_directory=VECTOR_DB_PATH
    )

    retriever = vectordb.as_retriever(search_kwargs={"k": 4})

    generator = pipeline(
        "text2text-generation",
        model=LLM_MODEL,
        tokenizer=LLM_MODEL,
        max_new_tokens=200,
        temperature=0.1,
        top_p=0.9,
        repetition_penalty=1.2
    )

    llm = HuggingFacePipeline(pipeline=generator)

    prompt = PromptTemplate(
        input_variables=["context", "question"],
        template="""
You are AskSASTRA, the official SASTRA University assistant.
Answer strictly from the given context.
If information is missing, say INSUFFICIENT_DATA.

Context:
{context}

Question:
{question}

Answer:
"""
    )

    qa_chain = RetrievalQA.from_chain_type(
        llm=llm,
        retriever=retriever,
        chain_type="stuff",
        chain_type_kwargs={"prompt": prompt},
        return_source_documents=False
    )

    print("✅ AskSASTRA initialized")


initialize_model()

# ============================================================
# ROUTES
# ============================================================

@app.route("/", methods=["GET"])
def home():
    return jsonify({"message": "AskSASTRA API running 🚀"})


@app.route("/chat", methods=["POST"])
def chat():
    data = request.get_json(force=True)
    query = data.get("query", "").strip()
    lang = data.get("language", "en")

    if not query:
        return jsonify({"answer": "Please ask a valid question."})

    # Translate input → English
    if lang != "en":
        try:
            query_en = GoogleTranslator(source=lang, target="en").translate(query)
        except:
            query_en = query
    else:
        query_en = query

    # RAG inference
    try:
        result = qa_chain.invoke({"query": query_en})
        answer_en = clean_llm_output(result.get("result", ""))
    except:
        answer_en = ""

    if not answer_en:
        answer_en = (
            "I could not find confident information for this question. "
            "Please visit the official SASTRA website."
        )

    # URL handling
    extracted_urls = extract_urls(answer_en)
    all_urls = list(set(extracted_urls + MANDATORY_URLS))

    # Translate output back
    if lang != "en":
        try:
            final_answer = GoogleTranslator(source="en", target=lang).translate(answer_en)
        except:
            final_answer = answer_en
    else:
        final_answer = answer_en

    return jsonify({
        "answer": final_answer,
        "urls": all_urls
    })


@app.route("/retrain", methods=["POST"])
def retrain():
    data = request.get_json(force=True)
    context = data.get("context", "").strip()

    if not context:
        return jsonify({"error": "Context text required"}), 400

    doc = Document(
        page_content=context,
        metadata={"source": "user_retrain"}
    )

    splitter = RecursiveCharacterTextSplitter(
        chunk_size=600,
        chunk_overlap=50
    )
    chunks = splitter.split_documents([doc])

    vectordb.add_documents(chunks)
    vectordb.persist()
    initialize_model()

    return jsonify({
        "status": "success",
        "message": "Successfully retrained with your data"
    })