pdf_bot.py

import os
from dotenv import load_dotenv
from langchain_community.document_loaders import PyPDFLoader
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import RetrievalQA
from langchain_core.language_models import BaseChatModel
from langchain_core.outputs import ChatResult, ChatGeneration
from langchain_core.messages import AIMessage
from groq import Groq

load_dotenv()
groq_api_key = os.getenv("GROQ_API_KEY")

# ✅ Custom wrapper
class ChatGroq(BaseChatModel):
    model: str = "llama3-8b-8192"
    temperature: float = 0.3
    groq_api_key: str = None

    def __init__(self, model="llama3-8b-8192", temperature=0.3, api_key=None):
        super().__init__()
        self.model = model
        self.temperature = temperature
        self.groq_api_key = api_key
        self._client = Groq(api_key=api_key)  # ✅ use _client instead of client

    def _generate(self, messages, stop=None):
        prompt = [{"role": "user", "content": self._get_message_text(messages)}]
        response = self._client.chat.completions.create(
            model=self.model,
            messages=prompt,
            temperature=self.temperature,
            max_tokens=1024
        )
        content = response.choices[0].message.content.strip()
        return ChatResult(generations=[ChatGeneration(message=AIMessage(content=content))])

    def _get_message_text(self, messages):
        if isinstance(messages, list):
            return " ".join([msg.content for msg in messages])
        return messages.content

    @property
    def _llm_type(self) -> str:
        return "chat-groq"

# ✅ Function to return a QA chain
def create_qa_chain_from_pdf(pdf_path):
    loader = PyPDFLoader(pdf_path)
    documents = loader.load()
    
    splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    texts = splitter.split_documents(documents)

    embeddings = HuggingFaceEmbeddings(model_name="BAAI/bge-m3")
    vectorstore = FAISS.from_documents(texts, embeddings)

    llm = ChatGroq(model="llama3-8b-8192", temperature=0.3, api_key=groq_api_key)

    qa_chain = RetrievalQA.from_chain_type(
        llm=llm,
        chain_type="stuff",
        retriever=vectorstore.as_retriever(search_kwargs={"k": 1}),
        return_source_documents=True
    )
    return qa_chain