The code helps you query your own personal directory in your pc

documentbot.py

from langchain.document_loaders import DirectoryLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings import HuggingFaceInstructEmbeddings
from langchain.vectorstores import FAISS
from transformers import T5Tokenizer, T5ForConditionalGeneration
import speech_recognition as sr
import sys

def get_context_faiss(query):
    search_results = db.similarity_search(query)
    context_doc = ' '.join([ doc.page_content for doc in search_results])
    return context_doc

if __name__ == "__main__":
    documents = []

    # Mention here your own custom path
    path = "./test/"
    for document in ["pdf", "md", "txt", "csv"]:
        loader = DirectoryLoader(path, glob=f'*.{document}')
        documents.extend(loader.load())

    text_splitter = RecursiveCharacterTextSplitter(chunk_size = 500, chunk_overlap = 0)
    all_splits = text_splitter.split_documents(documents)

    embeddings = HuggingFaceInstructEmbeddings(
            model_name="hkunlp/instructor-large",
            model_kwargs={"device": "cpu"}
        )
    vectorstore = FAISS.from_documents(all_splits, embeddings)
    vectorstore.save_local("vector_store/faiss_index_local_data")
    db = FAISS.load_local("vector_store/faiss_index_local_data", embeddings)

    parameter = sys.argv[-1]
    while True:
        if parameter == "speech":
            r = sr.Recognizer()
            with sr.Microphone() as source:
                print("Say something!")
                audio = r.listen(source)
            query = r.recognize_whisper(audio, language="english")
        else:
            query = input("Enter your query: ")
        print(f"Query is : {query}")
        if query == 'quit':
            break

        context = get_context_faiss(query)
        prompt_txt = f"""{context}
        Q: {query}
        A: 
        """
        
        ## In the current example we are using Flan-T5-llm to get our answers
        tokenizer = T5Tokenizer.from_pretrained("google/flan-t5-xxl")
        model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-xxl")

        input_ids = tokenizer(prompt_txt, return_tensors="pt").input_ids

        outputs = model.generate(input_ids)
        print(tokenizer.decode(outputs[0]))