app.py

import streamlit as st
from langchain.memory import ConversationBufferMemory
from llama_index.core.indices.query.schema import QueryBundle
from llama_index.core import Document, VectorStoreIndex
from llama_index.core.text_splitter import SentenceSplitter
from llama_index.core.retrievers import QueryFusionRetriever
from llama_index.retrievers.bm25 import BM25Retriever
from llama_index.core.postprocessor import SentenceTransformerRerank
from llama_index.core.prompts import PromptTemplate
from llama_index.core.query_engine import RetrieverQueryEngine
from llama_index.embeddings.gemini import GeminiEmbedding
from llama_index.llms.gemini import Gemini
from llama_index.core import Settings
from llama_index.vector_stores.faiss import FaissVectorStore
from llama_index.core import (
    SimpleDirectoryReader,
    load_index_from_storage,
    VectorStoreIndex,
    StorageContext,
)
from llama_index.core.node_parser import SemanticSplitterNodeParser

import os
import faiss
import pickle
import spacy

# Load NLP model
# nlp = spacy.load("en_core_web_sm")

GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')

# Function to load documents
def load_documents(filename="documents.pkl"):
    with open(filename, "rb") as file:
        return pickle.load(file)

# Load stored documents
loaded_docs = load_documents()

# Function to split text into sentences
# def spacy_sentence_splitter(text):
#     doc = nlp(text)
#     return [sent.text for sent in doc.sents]
embed_model = GeminiEmbedding(model_name="models/embedding-001", use_async=False)
splitter = SemanticSplitterNodeParser(
    buffer_size=5, breakpoint_percentile_threshold=95, embed_model=embed_model
)
# splitter = SentenceSplitter(chunk_size=512, chunk_overlap=50, separator="\n")
nodes = splitter.get_nodes_from_documents([doc for doc in loaded_docs])
chunked_documents = [Document(text=node.text, metadata=node.metadata) for node in nodes]
# Process documents
# chunked_documents = [
#     Document(text=chunk_text, metadata=doc.metadata)
#     for doc in loaded_docs for chunk_text in spacy_sentence_splitter(doc.text)
# ]

# Configure LLM and embeddings
Settings.llm = Gemini(model="models/gemini-2.0-flash", api_key=GOOGLE_API_KEY, temperature=0.5)

dimension = 768
faiss_index = faiss.IndexFlatL2(dimension)
vector_store = FaissVectorStore(faiss_index=faiss_index)
storage_context = StorageContext.from_defaults(vector_store=vector_store)

# Build index
index = VectorStoreIndex.from_documents(
    documents=chunked_documents,
    storage_context=storage_context,
    embed_model=embed_model,
    show_progress=True
)
index.storage_context.persist()

# Initialize memory
memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)

def get_chat_history():
    return memory.load_memory_variables({})["chat_history"]

# Define chatbot prompt template
prompt_template = PromptTemplate(
    """You are a friendly college counselor with expertise in Indian technical institutes.
    Previous conversation context (if any):\n{chat_history}\n\n
    Available college information:\n{context_str}\n\n"
    User query: {query_str}\n\n
    Instructions:\n
    1. Provide a brief, direct answer using only the information available above\n
    2. If specific data is not available, clearly state that\n
    3. Keep responses under 3 sentences when possible\n
    4. If comparing colleges, use bullet points for clarity\n
    5. Use a friendly, conversational tone\n
    6. Always be interactive and ask follow-up questions\n
    7. Always try to give answers in points each point should focus on single aspect of the response.\n
    8. Always try to give conclusion of your answer in the end for the user to take a decision.\n
    Response:"""
)

# Configure retrieval and query engine
vector_retriever = index.as_retriever(similarity_top_k=10)
bm25_retriever = BM25Retriever.from_defaults(index=index, similarity_top_k=10)
hybrid_retriever = QueryFusionRetriever(
    [vector_retriever, bm25_retriever],
    similarity_top_k=10,
    num_queries=10,
    mode="reciprocal_rerank",
    use_async=False
)

reranker = SentenceTransformerRerank(
    model="cross-encoder/ms-marco-MiniLM-L-2-v2",
    top_n=10,
)

query_engine = RetrieverQueryEngine.from_args(
    retriever=hybrid_retriever,
    node_postprocessors=[reranker],
    llm=Settings.llm,
    verbose=True,
    prompt_template=prompt_template,
    use_async=False,
)

# Streamlit UI
st.title("📚 Precollege Chatbot")
st.write("Ask me anything about different colleges and their courses!")

# Custom CSS for WhatsApp-like interface
st.markdown("""
<style>
body {
    background-color: #111b21;
    color: #e9edef;
}
.stApp {
    background-color: #111b21;
}
.chat-container {
    padding: 10px;
    color: #111b21;
}
.user-message {
    background-color: #005c4b;
    color: #e9edef;
    padding: 10px 15px;
    border-radius: 15px;
    margin: 5px 0;
    max-width: 70%;
    margin-left: auto;
    margin-right: 10px;
}
.ai-message {
    background-color: #1f2c33;
    color: #e9edef;
    padding: 10px 15px;
    border-radius: 15px;
    margin: 5px 0;
    max-width: 70%;
    margin-right: auto;
    margin-left: 10px;
    box-shadow: 0 1px 2px rgba(255,255,255,0.1);
}
.ai-message table {
    border-collapse: collapse;
    width: 100%;
    margin: 10px 0;
}
.ai-message th, .ai-message td {
    border: 1px solid #e9edef;
    padding: 8px;
    text-align: left;
}
.ai-message th {
    background-color: #2a3942;
}
.message-container {
    display: flex;
    margin-bottom: 10px;
}
.stTextInput input {
    border-radius: 20px;
    padding: 10px 20px;
    border: 1px solid #ccc;
    background-color: #2a3942;
    color: #e9edef;
}
.stButton button {
    border-radius: 50%;  /* Make it circular */
    width: 40px;
    height: 40px;
    padding: 0px;
    background-color: #005c4b;
    color: #e9edef;
    font-size: 20px;
    display: flex;
    align-items: center;
    justify-content: center;
    border: none;
    cursor: pointer;
}
.stButton button:hover {
    background-color: #00735e;
}
div[data-testid="stToolbar"] {
    display: none;
}
.stMarkdown {
    color: #e9edef;
}
header {
    background-color: #202c33 !important;
}
.ai-message table.ai-table {
    border-collapse: collapse;
    width: 100%;
    margin: 10px 0;
    background-color: #2a3942;
}

.ai-message table.ai-table th,
.ai-message table.ai-table td {
    border: 1px solid #e9edef;
    padding: 8px;
    text-align: left;
    color: #e9edef;
}

.ai-message table.ai-table th {
    background-color: #005c4b;
    font-weight: bold;
}

.ai-message table.ai-table tr:nth-child(even) {
    background-color: #1f2c33;
}
</style>
""", unsafe_allow_html=True)

if "chat_history" not in st.session_state:
    st.session_state.chat_history = []

# Create a container for chat messages
chat_container = st.container()

# Create a form for input
with st.form(key="message_form", clear_on_submit=True):
    col1, col2 = st.columns([5,1])
    with col1:
        user_input = st.text_input("", placeholder="Type a message...", label_visibility="collapsed")
    with col2:
        submit_button = st.form_submit_button("➤")

    if submit_button and user_input.strip():
        chat_history = get_chat_history()
        query_bundle = QueryBundle(query_str=f"{chat_history}\n\nUser: {user_input}")
        response_obj = query_engine.query(query_bundle)
        response_text = str(response_obj.response) if hasattr(response_obj, "response") else str(response_obj)
        
        memory.save_context({"query_str": user_input}, {"response": response_text})
        st.session_state.chat_history.append(("You", user_input))
        st.session_state.chat_history.append(("AI", response_text))

# Display chat history with custom styling
with chat_container:
    for role, message in st.session_state.chat_history:
        message = message.replace("</div>", "").replace("<div>", "")  # Sanitize the message
        if role == "You":
            st.markdown(
                f'<div class="message-container"><div class="user-message">{message}</div></div>',
                unsafe_allow_html=True
            )
        else:
            # Convert markdown tables to HTML tables with proper styling
            if "|" in message and "-|-" in message:  # Detect markdown tables
                # Split the message into lines
                lines = message.split("\n")
                table_html = []
                in_table = False
                formatted_lines = []
                
                for line in lines:
                    if "|" in line:
                        if not in_table:
                            in_table = True
                            table_html.append('<table class="ai-table">')
                            # Add header
                            header = line.strip().strip("|").split("|")
                            table_html.append("<tr>")
                            for h in header:
                                table_html.append(f"<th>{h.strip()}</th>")
                            table_html.append("</tr>")
                        elif "-|-" not in line:  # Skip separator line
                            # Add row
                            row = line.strip().strip("|").split("|")
                            table_html.append("<tr>")
                            for cell in row:
                                table_html.append(f"<td>{cell.strip()}</td>")
                            table_html.append("</tr>")
                    else:
                        if in_table:
                            in_table = False
                            table_html.append("</table>")
                            formatted_lines.append("".join(table_html))
                            table_html = []
                        formatted_lines.append(line)
                
                if in_table:
                    table_html.append("</table>")
                    formatted_lines.append("".join(table_html))
                
                message = "\n".join(formatted_lines)
            
            st.markdown(
                f'<div class="message-container"><div class="ai-message">{message}</div></div>',
                unsafe_allow_html=True
            )