app.py

from omegaconf import OmegaConf
import streamlit as st
import os
from PIL import Image
import re
from translate import Translator
from pydantic import Field
import sys

from llama_index.indices.managed.vectara import VectaraIndex
from llama_index.core.agent import ReActAgent
from llama_index.llms.openai import OpenAI
from llama_index.core.tools import QueryEngineTool, ToolMetadata
from llama_index.core.utils import print_text
from llama_index.core.agent.react.formatter import ReActChatFormatter
from llama_index.core.tools import FunctionTool

from prompts import prompt_template

learning_styles = ['traditional', 'Inquiry-based', 'Socratic']
languages = {'English': 'en', 'Spanish': 'es', 'French': 'fr', 'German': 'de', 'Arabic': 'ar', 'Chinese': 'zh-cn', 
             'Hebrew': 'he', 'Hindi': 'hi', 'Italian': 'it', 'Japanese': 'ja', 'Korean': 'ko', 'Portuguese': 'pt'}
initial_prompt = "How can I help you today?"


def launch_bot():
    def reset():
        cfg = st.session_state.cfg
        llm = OpenAI(model="gpt-4o", temperature=0)        
        tr_prompt = Translator(to_lang=languages[cfg.language]).translate(initial_prompt)
        st.session_state.messages = [{"role": "assistant", "content": tr_prompt, "avatar": "🦖"}]
        st.session_state.thinking_prompt = Translator(to_lang=languages[cfg.language]).translate("Thinking...")
        vectara = VectaraIndex(vectara_api_key=cfg.api_key, 
                               vectara_customer_id=cfg.customer_id, 
                               vectara_corpus_id=cfg.corpus_id)

        # Create tool to adapt output to style, age and language
        def adjust_response_to_student(
                text: str = Field(descrition='the original text'), 
                age: int = Field(description='the age of the student. An integer'), 
                style: str = Field(description='teaching style'), 
                language: str = Field(description='the language')
            ) -> str:
            """
            Rephrase the text to match the student's age, desired teaching style and language
            """
            llm = OpenAI(model="gpt-4o", temperature=0)
            print(f"DEBUG: Adjusting response to student age {age}, style {style} and language {language}")
            prompt = f'''
            The following is response the teacher is planning to provide to a student based on their question.
            Please adjust the response to match the student's age of {age}, the {style} teaching style.
            For example, in the inquiry-based teaching style, choose to ask questions that encourage the student to think critically instead of repsonding directly with the answer.
            Or in the socratic teaching style, choose to ask questions that lead the student to the answer.
            Always respond in the {language} language.
            original response: {text}
            adjusted response:
            '''
            response = llm.complete(prompt)
            return response


        # Create the Vectara Tool
        vectara_tool = QueryEngineTool(
            query_engine = vectara.as_query_engine(summary_enabled = True, summary_num_results = 10, summary_response_lang = languages[cfg.language],
                                                   summary_prompt_name = "vectara-summary-ext-24-05-large",
                                                   vectara_query_mode = "mmr", rerank_k = 50, mmr_diversity_bias = 0.1,
                                                   n_sentence_before = 5, n_sentence_after = 5),
            metadata = ToolMetadata(name="vectara", 
                                    description="""
                                    A tool that is able to answer questions about the justice, morality, politics and related topics.
                                    Based on transcripts of recordings from the Justice Harvard class that includes a lot of content on these topics.
                                    When using the tool it's best to ask simple short questions.
                                    """),
        )

        rephrase_tool = FunctionTool.from_defaults(adjust_response_to_student)

        # Create the agent
        prompt = prompt_template.replace("{style}", cfg.style) \
                                .replace("{language}", cfg.language) \
                                .replace("{student_age}", str(cfg.student_age))
        
        st.session_state.agent = ReActAgent.from_tools(
            tools=[vectara_tool, rephrase_tool], llm=llm, 
            verbose=True,
            react_chat_formatter = ReActChatFormatter(system_header=prompt)
        )


    if 'cfg' not in st.session_state:
        cfg = OmegaConf.create({
            'customer_id': str(os.environ['VECTARA_CUSTOMER_ID']),
            'corpus_id': str(os.environ['VECTARA_CORPUS_ID']),
            'api_key': str(os.environ['VECTARA_API_KEY']),
            'style': learning_styles[0],
            'language': 'English',
            'student_age': 18
        })
        st.session_state.cfg = cfg
        st.session_state.style = cfg.style
        st.session_state.language = cfg.language
        st.session_state.student_age = cfg.student_age
        reset()

    cfg = st.session_state.cfg
    st.set_page_config(page_title="Teaching Assistant", layout="wide")

    # left side content
    with st.sidebar:
        image = Image.open('Vectara-logo.png')
        st.image(image, width=250)
        st.markdown("## Welcome to the Justice Harvard e-learning assistant demo.\n\n\n")

        st.markdown("\n")
        cfg.style = st.selectbox('Learning Style:', learning_styles)
        if st.session_state.style != cfg.style:
            st.session_state.style = cfg.style
            reset()

        st.markdown("\n")
        cfg.language = st.selectbox('Language:', languages.keys())
        if st.session_state.language != cfg.language:
            st.session_state.langage = cfg.language
            reset()

        st.markdown("\n") 
        cfg.student_age = st.number_input(
            'Student age:',  min_value=13, value=cfg.student_age,
            step=1, format='%i'
        )
        if st.session_state.student_age != cfg.student_age:
            st.session_state.student_age = cfg.student_age
            reset()

        st.markdown("\n\n")
        if st.button('Start Over'):
            reset()

        st.markdown("---")
        st.markdown(
            "## How this works?\n"
            "This app was built with [Vectara](https://vectara.com).\n\n"
            "It demonstrates the use of Agentic Chat functionality with Vectara"
        )
        st.markdown("---")


    if "messages" not in st.session_state.keys():
        reset()

    # Display chat messages
    for message in st.session_state.messages:
        with st.chat_message(message["role"], avatar=message["avatar"]):
            st.write(message["content"])

    # User-provided prompt
    if prompt := st.chat_input():
        st.session_state.messages.append({"role": "user", "content": prompt, "avatar": '🧑‍💻'})
        with st.chat_message("user", avatar='🧑‍💻'):
            print_text(f"Starting new question: {prompt}\n", color='green')
            st.write(prompt)

    # Generate a new response if last message is not from assistant
    if st.session_state.messages[-1]["role"] != "assistant":
        with st.chat_message("assistant", avatar='🤖'):
            with st.spinner(st.session_state.thinking_prompt):
                res = st.session_state.agent.chat(prompt)
                cleaned = re.sub(r'\[\d+\]', '', res.response)
                st.write(cleaned)
            message = {"role": "assistant", "content": cleaned, "avatar": '🤖'}
            st.session_state.messages.append(message)
    
    sys.stdout.flush()

if __name__ == "__main__":
    launch_bot()