app.py

import os
import re
from pprint import pprint
from pypdf import PdfReader
from langchain.text_splitter import RecursiveCharacterTextSplitter, SentenceTransformersTokenTextSplitter
import chromadb
import openai
from openai import OpenAI
import gradio as gr

# Setup environment variables and OpenAI API key
os.environ['OPENAI_API_KEY'] = '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'
openai.api_key = os.environ["OPENAI_API_KEY"]


# Load and preprocess the PDF document
ipcc_report_file = "IPCC_AR6_WGII_TechnicalSummary (1).pdf"
reader = PdfReader(ipcc_report_file)
ipcc_texts = [page.extract_text().strip() for page in reader.pages]

# Filter out beginning and end of document
ipcc_texts_filt = ipcc_texts[5: -5]

# Remove all header / footer texts
ipcc_wo_header_footer = [re.sub(r'\d+\nTechnical Summary', '', s) for s in ipcc_texts_filt]
ipcc_wo_header_footer = [re.sub(r'\nTS', '', s) for s in ipcc_wo_header_footer]
ipcc_wo_header_footer = [re.sub(r'TS\n', '', s) for s in ipcc_wo_header_footer]

# Split the document text into chunks
char_splitter = RecursiveCharacterTextSplitter(
    separators=["\n\n", "\n", ". ", " ", ""],
    chunk_size=1000,
    chunk_overlap=0.2
)
texts_char_splitted = char_splitter.split_text('\n\n'.join(ipcc_wo_header_footer))

# Split text into tokens
token_splitter = SentenceTransformersTokenTextSplitter(
    chunk_overlap=0.2,
    tokens_per_chunk=256
)
texts_token_splitted = []
for text in texts_char_splitted:
    try:
        texts_token_splitted.extend(token_splitter.split_text(text))
    except Exception as e:
        print(f"Error in text: {text}, {e}")
        continue

# Create and populate the vector database
chroma_client = chromadb.PersistentClient(path="db")
chroma_collection = chroma_client.get_or_create_collection("ipcc")
ids = [str(i) for i in range(len(texts_token_splitted))]
chroma_collection.add(
    ids=ids,
    documents=texts_token_splitted
)

# Define the rag function based on the notebook content
def rag(query, n_results=5):
    res = chroma_collection.query(query_texts=[query], n_results=n_results)
    docs = res["documents"][0]
    joined_information = ';'.join([f'{doc}' for doc in docs])
    messages = [
        {
            "role": "system",
            "content": "You are a helpful expert on climate change. Your users are asking questions about information contained in attached information."
            "You will be shown the user's question, and the relevant information. Answer the user's question using only this information."
        },
        {"role": "user", "content": f"Question: {query}. \n Information: {joined_information}"}
    ]
    openai_client = OpenAI()
    model = "gpt-3.5-turbo"
    response = openai_client.chat.completions.create(
        model=model,
        messages=messages,
    )
    content = response.choices[0].message.content
    return content

# Create Gradio interface
iface = gr.Interface(
    fn=rag,
    inputs=["text"],
    outputs="text",
    title="Climate Change RAG (Using OpenAI)",
    description="Ask questions about the impact of climate change and get answers based on the provided document."
)

# Launch the interface
iface.launch(debug = True)