document_chat.py

import os
from langchain.vectorstores import Chroma
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.document_loaders import PyMuPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
from langchain.llms import HuggingFaceHub

# Constants
CHROMA_DB_PATH = "chroma_db"
SENTENCE_TRANSFORMER_MODEL = "sentence-transformers/all-MiniLM-L6-v2"  # Corrected model name
LLM_MODEL = "HuggingFaceH4/zephyr-7b-beta"  # Free chatbot model from Hugging Face

# Initialize vector store
def initialize_vector_store():
    """Initialize or load ChromaDB vector store"""
    embeddings = HuggingFaceEmbeddings(model_name=SENTENCE_TRANSFORMER_MODEL)
    vector_store = Chroma(persist_directory=CHROMA_DB_PATH, embedding_function=embeddings)
    return vector_store

vector_store = initialize_vector_store()

def ingest_pdf(pdf_path):
    """Processes a PDF, splits text, and stores embeddings in ChromaDB."""
    loader = PyMuPDFLoader(pdf_path)
    documents = loader.load()

    # Split text into smaller chunks
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
    split_docs = text_splitter.split_documents(documents)

    # Store in vector database
    vector_store.add_documents(split_docs)
    vector_store.persist()

def process_query_with_memory(query, chat_history=[]):
    """Retrieves relevant document chunks and generates a conversational response."""
    retriever = vector_store.as_retriever()

    # Initialize chat memory
    memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)

    # Load a free Hugging Face model
    llm = HuggingFaceHub(repo_id=LLM_MODEL, model_kwargs={"max_new_tokens": 500})

    # Create a conversational retrieval chain
    qa_chain = ConversationalRetrievalChain(
        llm=llm,
        retriever=retriever,
        memory=memory
    )

    return qa_chain.run({"question": query, "chat_history": chat_history})