categories/accomodation/__init__.py

from .model import InformationExtractedFromABillReceipt as PydanticModel

from langchain.chains import LLMChain
from langchain.chat_models import ChatOpenAI
from langchain.output_parsers import PydanticOutputParser, OutputFixingParser
from langchain.prompts import (
    ChatPromptTemplate,
    HumanMessagePromptTemplate,
    SystemMessagePromptTemplate,
)

model = ChatOpenAI(
    temperature=0.6,
    max_tokens=300,
    n=1,
    request_timeout=None,
    model_kwargs={
        'stop': None,
        'top_p': 1,
    }
)

# Build category chain
system_message_prompt = SystemMessagePromptTemplate.from_template(
    "You are tasked with developing an OCR data extraction system for hotel bills in PDF "
    "format given as text. The system should extract important information necessary for "
    "the reimbursement process from a college. Your prompt should fetch the following "
    "essential details from the hotel bill: hotel name, address, bill number/invoice "
    "number, booking ID / confirmation ID / booking number, check-in date and time, "
    "check-out date and time, total amount, booking platform, bill date.\n"
    "Ensure that the system accurately extracts the above information from the OCR text "
    "of the hotel bill. Fields with formats specified as date, time, or datetime should "
    "be ISO 8601 compliant.\n"
    "{format_instructions}"
)
human_message_prompt = HumanMessagePromptTemplate.from_template("{text}")
chat_prompt = ChatPromptTemplate.from_messages(
    [system_message_prompt, human_message_prompt]
)
output_parser = PydanticOutputParser(pydantic_object=PydanticModel)
fixing_parser = OutputFixingParser.from_llm(llm=model, parser=output_parser)
chain = LLMChain(llm=model, prompt=chat_prompt, output_parser=fixing_parser)