import pandas as pd
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np

faq_df = pd.read_csv("data/koda_faq.csv")
faq_questions = faq_df["question"].tolist()
faq_answers = faq_df["answer"].tolist()

embedder = SentenceTransformer("all-MiniLM-L6-v2")
faq_embs = embedder.encode(faq_questions, normalize_embeddings=True)

def retrieve_answer(user_q, top_k=1, thresh=0.35):
    q_emb = embedder.encode([user_q], normalize_embeddings=True)
    sims = cosine_similarity(q_emb, faq_embs)[0]
    idx = int(np.argmax(sims))
    if sims[idx] >= thresh:
        return faq_answers[idx], float(sims[idx])
    return None, float(sims[idx])

def koda_assistant(user_input, history):
    # Check if input matches FAQ keywords
    for key, answer in FAQ.items():
        if key in user_input.lower():
            history.append(("👤 " + user_input, "🤖 " + answer))
            return history, history

    # Fallback: generic QA pipeline (using a sample context)
    context = "Koda Appliance Assistant helps with appliance troubleshooting and leasing information."
    try:
        result = qa_pipeline(question=user_input, context=context)
        answer = result["answer"]
    except Exception:
        answer = "I'm not sure about that, but our support team can help."
    
    history.append(("👤 " + user_input, "🤖 " + answer))
    return history, history

with gr.Blocks(title="Koda Appliance Assistant") as demo:
    gr.Markdown("# ⚙️ Koda Appliance Assistant")
    gr.Markdown("Ask me about appliance issues or Koda leasing FAQs.")

    chatbot = gr.Chatbot(height=400)
    user_input = gr.Textbox(placeholder="Type your question here...")
    state = gr.State([])

    submit_btn = gr.Button("Ask")

    submit_btn.click(fn=koda_assistant, inputs=[user_input, state], outputs=[chatbot, state])
    user_input.submit(fn=koda_assistant, inputs=[user_input, state], outputs=[chatbot, state])

# Expose app object
app = demo

if __name__ == "__main__":
    demo.launch()