Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Imports
|
2 |
+
import streamlit as st
|
3 |
+
from PyPDF2 import PdfReader
|
4 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
5 |
+
import os
|
6 |
+
from langchain_google_genai import GoogleGenerativeAIEmbeddings
|
7 |
+
import google.generativeai as genai
|
8 |
+
from langchain_community.vectorstores import FAISS
|
9 |
+
from langchain_google_genai import ChatGoogleGenerativeAI
|
10 |
+
from langchain.chains.question_answering import load_qa_chain
|
11 |
+
from langchain.prompts import PromptTemplate
|
12 |
+
from dotenv import load_dotenv
|
13 |
+
|
14 |
+
# Load environment variables
|
15 |
+
load_dotenv()
|
16 |
+
os.getenv("GOOGLE_API_KEY")
|
17 |
+
genai.configure(api_key=os.getenv("GOOGLE_API_KEY")) # Configure Google Generative AI
|
18 |
+
|
19 |
+
# Extracts text from all pages of provided PDF documents
|
20 |
+
def get_pdf_text(pdf_docs):
|
21 |
+
text = ""
|
22 |
+
for pdf in pdf_docs:
|
23 |
+
pdf_reader = PdfReader(pdf)
|
24 |
+
for page in pdf_reader.pages:
|
25 |
+
text += page.extract_text()
|
26 |
+
return text
|
27 |
+
|
28 |
+
# Splits text into chunks of 10,000 characters with 1,000 character overlap
|
29 |
+
def get_text_chunks(text):
|
30 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=10000, chunk_overlap=1000)
|
31 |
+
chunks = text_splitter.split_text(text)
|
32 |
+
return chunks
|
33 |
+
|
34 |
+
# Creates and saves a FAISS vector store from text chunks
|
35 |
+
def get_vector_store(text_chunks):
|
36 |
+
embeddings = GoogleGenerativeAIEmbeddings(model="models/embedding-001")
|
37 |
+
vector_store = FAISS.from_texts(text_chunks, embedding=embeddings)
|
38 |
+
vector_store.save_local("faiss_index")
|
39 |
+
|
40 |
+
# Creates and returns a conversational chain for question answering
|
41 |
+
def get_conversational_chain():
|
42 |
+
prompt_template = """
|
43 |
+
Answer the question concisely, focusing on the most relevant and important details from the PDF context.
|
44 |
+
Refrain from mentioning any mathematical equations, even if they are present in provided context.
|
45 |
+
Focus on the textual information available. Please provide direct quotations or references from PDF
|
46 |
+
to back up your response. If the answer is not found within the PDF,
|
47 |
+
please state "answer is not available in the context.\n\n
|
48 |
+
|
49 |
+
Context:\n {context}?\n
|
50 |
+
Question: \n{question}\n
|
51 |
+
|
52 |
+
Example response format:
|
53 |
+
- Overview: (brief summary or introduction)
|
54 |
+
- Key points:
|
55 |
+
(point 1: paragraph for main details)
|
56 |
+
(point 2: paragraph for main details)
|
57 |
+
...
|
58 |
+
|
59 |
+
Use a mix of paragraphs and points to effectively convey the information.
|
60 |
+
"""
|
61 |
+
|
62 |
+
# Adjust temperature parameter to lower value to:
|
63 |
+
# reduce model creativity & focus on factual accuracy
|
64 |
+
model = ChatGoogleGenerativeAI(model="gemini-pro", temperature=0.2)
|
65 |
+
|
66 |
+
prompt = PromptTemplate(template=prompt_template, input_variables=["context", "question"])
|
67 |
+
chain = load_qa_chain(model, chain_type="stuff", prompt=prompt)
|
68 |
+
|
69 |
+
return chain
|
70 |
+
|
71 |
+
# Processes user question and provides a response
|
72 |
+
def user_input(user_question):
|
73 |
+
embeddings = GoogleGenerativeAIEmbeddings(model="models/embedding-001")
|
74 |
+
|
75 |
+
new_db = FAISS.load_local("faiss_index", embeddings)
|
76 |
+
docs = new_db.similarity_search(user_question)
|
77 |
+
|
78 |
+
chain = get_conversational_chain()
|
79 |
+
|
80 |
+
response = chain.invoke(
|
81 |
+
{"input_documents": docs, "question": user_question},
|
82 |
+
return_only_outputs=True
|
83 |
+
)
|
84 |
+
|
85 |
+
st.write("Reply: ", response["output_text"])
|
86 |
+
|
87 |
+
# Streamlit UI
|
88 |
+
def main():
|
89 |
+
st.set_page_config(page_title="Chat with PDFs", page_icon="")
|
90 |
+
st.header("Chat with multiple PDFs using AI 💬")
|
91 |
+
|
92 |
+
user_question = st.text_input("Ask a Question from PDF file(s)")
|
93 |
+
|
94 |
+
if user_question:
|
95 |
+
user_input(user_question)
|
96 |
+
|
97 |
+
with st.sidebar:
|
98 |
+
st.title("Menu ✨")
|
99 |
+
pdf_docs = st.file_uploader("Upload your PDF Files and Click on the Submit & Process Button ",
|
100 |
+
accept_multiple_files=True)
|
101 |
+
|
102 |
+
if st.button("Submit & Process"):
|
103 |
+
with st.spinner("Processing..."):
|
104 |
+
raw_text = get_pdf_text(pdf_docs)
|
105 |
+
text_chunks = get_text_chunks(raw_text)
|
106 |
+
get_vector_store(text_chunks)
|
107 |
+
st.success("Done ✨")
|
108 |
+
|
109 |
+
if __name__ == "__main__":
|
110 |
+
main()
|