app.py

from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM, AutoModelForSequenceClassification
import streamlit as st
from llama_cpp import Llama
from dotenv import load_dotenv
import os
import torch

load_dotenv()

try:
    REPO_ID = os.getenv('REPO_ID')
    INTENT_MODEL = os.getenv('INTENT_MODEL')
    CHITCHAT_MODEL = os.getenv('CHITCHAT_MODEL')

    if not all([REPO_ID, INTENT_MODEL, CHITCHAT_MODEL]):
        raise EnvironmentError("One or more required environment variables are missing.")

except Exception as e:
    st.error(f"Environment setup failed: {e}")
    raise SystemExit(e)

try:
    llm_hr = Llama.from_pretrained(
        repo_id=REPO_ID,
        filename="unsloth.F16.gguf"
    )
    intent_tokenizer = AutoTokenizer.from_pretrained(INTENT_MODEL)
    intent_model = AutoModelForSequenceClassification.from_pretrained(INTENT_MODEL)
    tokenizer = AutoTokenizer.from_pretrained(CHITCHAT_MODEL)
    model = AutoModelForSeq2SeqLM.from_pretrained(CHITCHAT_MODEL)
    pipe_chitchat = pipeline('text2text-generation', model=model, tokenizer=tokenizer)

except Exception as e:
    st.error(f"Model loading failed: {e}")
    raise SystemExit(e)


intent_labels = ["Employee Benefits & Policies", "Employee Support & Self-Service", "Recruitment & Onboarding"]
total_labels = ["😂 Let's Chat & Chill"] + intent_labels


st.set_page_config(page_title="AI HR Chatbot", layout="wide")
st.markdown("""
<style>
body { background-color: #f6f9fc; }
.sidebar .sidebar-content { background-color: #ffffff; }
.chat-container { max-height: 500px; overflow-y: auto; padding: 1rem; display: flex; flex-direction: column; }
.chat-box { display: inline-block; border-radius: 10px; padding: 8px 12px; margin: 5px 0; max-width: 70%; word-wrap: break-word; }
.user-msg { background-color: #708090; text-align: right; align-self: flex-end; color: white; }
.bot-msg { background-color: #d3d3d3; text-align: left; align-self: flex-start; color: black; }
div.title { font-size: 2rem; font-weight: bold; margin: 20px 0; color: #0077cc; }
</style>
""", unsafe_allow_html=True)

if "intent" not in st.session_state:
    st.session_state.intent = total_labels[0]
if "messages" not in st.session_state:
    st.session_state.messages = []


def select_intent(current_intent):
    display_labels = [f"✅ {intent}" if intent == current_intent else intent for intent in total_labels]
    selected = st.sidebar.radio(
        "💼 Select Your HR Intent",
        display_labels,
        index=display_labels.index(f"✅ {current_intent}") if f"✅ {current_intent}" in display_labels else 0
    )
    return selected.replace("✅ ", "")


def determine_intent(text):
    try:
        inputs = intent_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
        with torch.no_grad():
            logits = intent_model(**inputs).logits
        predicted_label = torch.argmax(logits, dim=1).item()
        return predicted_label
    except Exception as e:
        st.error(f"Intent determination failed: {e}")
        return 0  


def render_chat():
    if st.session_state.messages:
        st.markdown('<div class="chat-container">', unsafe_allow_html=True)
        for msg in st.session_state.messages:
            role_class = "user-msg" if msg["role"] == "user" else "bot-msg"
            with st.chat_message(msg["role"], avatar=msg.get("avatar")):
                st.markdown(f'<div class="chat-box {role_class}">{msg["content"]}</div>', unsafe_allow_html=True)
        st.markdown('</div>', unsafe_allow_html=True)


def generate_hr_response(user_prompt, intent):
    hr_prompt = """You are an HR Assistant at our company.
    Your role is to assist employees by providing accurate and concise responses regarding company policies and HR-related questions.

    ### Instruction:
    {}

    ### Input:
    {}

    ### Response:
    {}"""
    instruction = f"Answer the HR-related query categorized as '{intent}'."
    formatted_prompt = hr_prompt.format(instruction, user_prompt, "") + tokenizer.eos_token

    try:
        response = llm_hr.create_chat_completion(
            messages=[{"role": "user", "content": formatted_prompt}],
            max_tokens=100,
        )
        return response["choices"][0]["message"]["content"]
    except Exception as e:
        st.error(f"Failed to generate HR response: {e}")
        return "⚠️ Sorry, I couldn't process your HR request at the moment."


st.markdown('<div class="title">AI HR Chatbot </div>', unsafe_allow_html=True)
st.markdown("<h6 style='color:rgb(131, 123, 160);'>Your personal assistant for HR queries and support.</h6>", unsafe_allow_html=True)

st.session_state.intent = select_intent(st.session_state.intent)
render_chat()


try:
    if prompt := st.chat_input("Ask me anything related to HR or just chat casually..."):
        st.session_state.messages.append({"role": "user", "content": prompt, "avatar": "man.png"})
        with st.chat_message("user", avatar="man.png"):
            st.markdown(f'<div class="chat-box user-msg">{prompt}</div>', unsafe_allow_html=True)

        if st.session_state.intent == "😂 Let's Chat & Chill":
            with st.spinner("😄 Chitchatting..."):
                try:
                    result = pipe_chitchat(prompt)
                    reply = result[0]['generated_text'].strip() 
                except Exception as e:
                    st.error(f"Chitchat generation failed: {e}")
                    reply = "Oops, I couldn't come up with a witty reply! 😅"
        else:
            detected_intent = determine_intent(prompt)
            predicted_label = intent_labels[detected_intent]

            if predicted_label != st.session_state.intent:
                
                st.warning(f"🔁 Automatically switched to **{predicted_label}** based on your query.")

            with st.spinner("Generating response with specialized HR model..."):
                reply = generate_hr_response(prompt, predicted_label)

        st.session_state.messages.append({"role": "bot", "content": reply, "avatar": "robot.png"})
        with st.chat_message("bot", avatar="robot.png"):
            st.markdown(f'<div class="chat-box bot-msg">{reply}</div>', unsafe_allow_html=True)

except Exception as e:
    st.error(f"Something went wrong while processing your input: {e}")