app.py

from sentence_transformers import SentenceTransformer
from PyPDF2 import PdfReader
from uuid import uuid4
from uuid import UUID
import os
from astrapy.db import AstraDB
import gradio as gr
from dotenv import load_dotenv
load_dotenv() 
from openai import OpenAI
client = OpenAI()


# Initialization
db = AstraDB(
    token=os.environ["ASTRA_DB_APPLICATION_TOKEN"],
    api_endpoint=os.environ["ASTRA_DB_API_ENDPOINT"],
)

model = SentenceTransformer('BAAI/bge-base-en-v1.5')

def get_embeddings(text):
    embeddings_1 = model.encode(text, normalize_embeddings=True)
    return embeddings_1.tolist()
        
def query(ques, col):
    emb = get_embeddings(ques)
    results = col.vector_find(emb, limit=2, fields={"text", "$vector"})
    return results

def read_pdf(pdf_path):
    reader = PdfReader(pdf_path)
    pdf_content = ""
    for i in range(len(reader.pages)):
        pdf_content += reader.pages[i].extract_text()
    return pdf_content
    
def create_chunks(content):
    batch_size = 1000
    overlap_size = 100

    chunks = []

    for i in range(0, len(content), batch_size - overlap_size):
        chunk = content[i:i + batch_size]
        chunks.append(chunk)

    return chunks

def create_docs(chunks):
    documents = []
    for i in (range(len(chunks))) :
        mydict = {
            "_id" : i+1,
            "text" : chunks[i],
            "$vector" : get_embeddings(chunks[i])
            }
        documents.append(mydict)
    return documents

def create_and_insert_docs(docs):
    user_id = str(uuid4())
    col = db.create_collection(f"user_{UUID(user_id).hex}", dimension=768, metric="cosine")
    res = col.insert_many(docs, partial_failures_allowed=True)
    return col

def get_answer(context, query):
    response = client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages=[
        {
        "role": "system",
        "content": "You are a document mining bot. You will given a user query, and user context. You have to give the reply to the user's query if the query's answer is in the context. If it isn't you reply with \"I don't know\""
        },
        {
        "role": "user",
        "content": f"Context: {context}\nQuery: {query}"
        }
    ],
    temperature=0.3,
    max_tokens=2000,
    top_p=1,
    frequency_penalty=0,
    presence_penalty=0
    )
    return response.choices[0].message.content

def query(col, ques):
    q = get_embeddings(ques)
    results = col.vector_find(q, limit=2, fields={"text", "$vector"})
    context = ""
    for res in results:
        context += res['text'] + "\n"
    return get_answer(context, ques)

def delete_col(col):
    db.delete_collection(col.collection_name)

def pipeline(files, user_input):
    total_chunks = []
    for file in files:
        content = read_pdf(file.name)
        chunks = create_chunks(content)
        total_chunks.extend(chunks)
    docs = create_docs(total_chunks)
    try:
        col = create_and_insert_docs(docs)
        ans = query(col, user_input)
        delete_col(col)
    except Exception as e:
        print(e)
        return "Sorry, we can't query that document right now. Please try a different document."
    return ans 

with gr.Blocks() as demo:
    gr.Markdown("# Chatbot Demo using DataStax Astra DB and OpenAI")
    about_bot = """## About the bot
    We created this bot using [DataStax Astra DB](https://www.datastax.com/products/datastax-astra) to store the vectors, and [BAAI/bge-base-en-v1.5](https://huggingface.co/BAAI/bge-base-en-v1.5) model to create embeddings, and [OpenAI's GPT-3.5-turbo](https://platform.openai.com/docs/models) for collecting the closest vectors and creating a human-friendly response.
    You can upload your pdf documents and chat with them!
    """
    gr.Markdown(about_bot)
    with gr.Row():
        with gr.Column():
            files = gr.Files(label = "Upload PDF Files", file_types = ['.pdf'])
            user_input = gr.Textbox(label = "Enter Query")
        with gr.Column():
            output = gr.Textbox(label = "Chatbot Response")
    with gr.Row():
        btn = gr.Button("Submit")

    btn.click(fn = pipeline, inputs=[files, user_input], outputs=output)

demo.launch()