import streamlit as st

import base64

from langchain.schema.messages import HumanMessage,AIMessage

from langchain_openai import ChatOpenAI

from PIL import Image

chain=ChatOpenAI(model="gpt-4-vision-preview",max_tokens=1024)

def encode_image(upload_file):

image_bytes=upload_file.getvalue()

base64_image=base64.b64encode(image_bytes).decode("utf-8")

return base64_image

def get_response(b64image,qsn):

msg=chain.invoke(

[

AIMessage(

content="you are a useful and intelligent bot who is very good at ocr related task , such getting insights from images of invoices"

),

HumanMessage(

content=[

{"type":"text","text":qsn},

{

"type": "image_url",

"image_url":f"data:image/jpg;base64,{b64image}"

}

]

)

]

)

return msg.content

# if "conversation" not in st.session_state:

# st.session_state.conversation=[]

def main():

st.title("CONVA INVOICE ANALYSIS SYSTEM")

upload_file=st.file_uploader("upload the invoice image..",type=["jpg"])

if upload_file is not None:

image=Image.open(upload_file)

st.image(image,caption="your uploaded invoice",use_column_width=True)

st.write("invoice image uploaded successfully")

b64_image=encode_image(upload_file)

st.success("image converted successfully")

user_question=st.text_input("ask anything related to the invoice")

submit_button=st.button("Submit")

if submit_button and user_question:

response=get_response(b64_image,user_question)

st.write(response)

if __name__=="__main__":

main()