app.py

import os
import streamlit as st
import time
from langchain.chat_models import ChatOpenAI
from langchain.chains import ConversationalRetrievalChain
from langchain.prompts import PromptTemplate
from langchain.memory import ConversationSummaryBufferMemory
from langchain.vectorstores import FAISS
from langchain.embeddings import OpenAIEmbeddings

# Set up the OpenAI API key
os.environ["OPENAI_API_KEY"] = os.environ.get("OPENAI_API_KEY")

# Load the FAISS index
embeddings = OpenAIEmbeddings()
vectorstore = FAISS.load_local("faiss_index", embeddings, allow_dangerous_deserialization=True)

# Create a retriever from the loaded vector store
retriever = vectorstore.as_retriever(search_kwargs={"k": 5})

# Define a prompt template for course recommendations
prompt_template = """
You are an AI course recommendation system. Your task is to recommend courses based on the user's description of their interests and goals, with a strong emphasis on matching the learning outcomes and syllabus content. Consider the summarized chat history to provide more relevant and personalized recommendations.
Summarized Chat History:
{chat_history}

User's Current Query: {query}
Based on the user's current query and chat history summary, here are some relevant courses from our database:
{context}

Please provide a personalized course recommendation. Your response should include:
1. A detailed explanation of how the recommended courses match the user's interests and previous queries, focusing primarily on the "What You Will Learn" section and the syllabus content.
2. A summary of each recommended course, highlighting:
   - The specific skills and knowledge the user will gain (from "What You Will Learn")
   - Key topics covered in the syllabus
   - Course level and language
   - The institution offering the course
3. Mention the course ratings if available.
4. Any additional advice or suggestions for the user's learning journey, based on the syllabus progression and their conversation history.
5. Provide the course URLs for easy access.

Prioritize courses that have the most relevant learning outcomes and syllabus content matching the user's description and previous interactions. If multiple courses are similarly relevant, you may suggest a learning path combining complementary courses.
Remember to be encouraging and supportive in your recommendation, and relate your suggestions to any preferences or constraints the user has mentioned in previous messages.
Recommendation:
"""

PROMPT = PromptTemplate(
    template=prompt_template,
    input_variables=["chat_history", "query", "context"]
)

# Initialize the language model
llm = ChatOpenAI(temperature=0.5, model_name="gpt-4-turbo")

# Set up conversation memory with summarization
memory = ConversationSummaryBufferMemory(llm=llm, max_token_limit=1000, memory_key="chat_history", return_messages=True)

# Create the conversational retrieval chain
qa_chain = ConversationalRetrievalChain.from_llm(
    llm=llm,
    retriever=retriever,
    memory=memory,
    combine_docs_chain_kwargs={"prompt": PROMPT}
)

# Streamlit app
st.set_page_config(page_title="AI Course Recommendation Chatbot", page_icon=":book:")
st.title("AI Course Recommendation Chatbot")

# Initialize chat history in session state
if "messages" not in st.session_state:
    st.session_state.messages = []

# Display chat messages from history on app rerun
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"])

# Accept user input
if prompt := st.chat_input("What are you looking to learn?"):
    # Add user message to chat history
    st.session_state.messages.append({"role": "user", "content": prompt})
    # Display user message in chat message container
    with st.chat_message("user"):
        st.markdown(prompt)

    # Assistant response generation with streaming effect
    with st.chat_message("assistant"):
        response = qa_chain({"query": prompt})
        response_text = response["response"]

        # Stream the response word by word
        for accumulated_response in response_text:
            st.markdown(accumulated_response, unsafe_allow_html=True)

    # Add assistant response to chat history
    st.session_state.messages.append({"role": "assistant", "content": response_text})